Image encoding / decoding method and apparatus, and recording medium storing bitstream

Abstract

A method and an apparatus, according to the present disclosure, may partition a current block into a plurality of partitions, generate a prediction block of the current block by performing prediction on each of the partitions, derive a residual block of the current block on the basis of an inverse transform for the current block, and reconstruct the current block on the basis of the prediction block and the residual block of the current block. Here, the step of deriving the residual block of the current block may comprise the steps of: deriving a virtual intra prediction mode for the current block; and determining a transform set for inverse transform on the basis of the virtual intra prediction mode.

Description

Video encoding / decoding method and device, and recording medium storing bitstream The present invention relates to a video encoding / decoding method and device, and a recording medium storing a bitstream. Recently, the demand for high-resolution, high-quality images, such as HD (High Definition) images and UHD (Ultra High Definition) images, is increasing in various application fields, and accordingly, high-efficiency image compression technologies are being discussed. There are various technologies for image compression, such as inter prediction technology that predicts pixel values ​​included in the current picture from pictures before or after the current picture, intra prediction technology that predicts pixel values ​​included in the current picture using pixel information within the current picture, and entropy encoding technology that assigns short codes to values ​​with high frequency of appearance and long codes to values ​​with low frequency of appearance, and these technologies can be used to effectively compress and transmit or store image data. The present disclosure provides a method and apparatus for performing a transformation using a non-separable primary transformation. The present disclosure provides a method and apparatus for performing a transformation using a non-separable first-order transform kernel of reduced dimensionality. The present disclosure provides a method and apparatus for determining a transformation set and / or a transformation kernel and signaling the same. A video decoding method and device according to the present disclosure can divide a current block into a plurality of partitions, perform prediction on each of the partitions to generate a prediction block of the current block, derive a residual block of the current block based on an inverse transformation for the current block, and reconstruct the current block based on the prediction block and the residual block of the current block. Here, the step of deriving the residual block of the current block can include the step of deriving a virtual intra prediction mode for the current block, and the step of determining a transform set for the inverse transformation based on the virtual intra prediction mode. In the video decoding method and device according to the present disclosure, the virtual intra prediction mode can be derived as a pre-defined fixed mode. In the image decoding method and device according to the present disclosure, the virtual intra prediction mode can be derived based on a division direction according to geometric division of the current block. In the image decoding method and device according to the present disclosure, the virtual intra prediction mode can be derived as an intra prediction mode that matches or is closest to a division direction according to geometric division of the current block. In the video decoding method and device according to the present disclosure, the virtual intra prediction mode can be derived as an intra prediction mode that matches or is closest to a direction perpendicular to a division direction according to geometric division of the current block. In the video decoding method and device according to the present disclosure, the virtual intra prediction mode can be derived based on at least one of a partition type of the current block or a partition index corresponding to the partition type. In the video decoding method and device according to the present disclosure, the virtual intra prediction mode may be derived as a decoder side intra mode derivation (DIMD) mode. Here, the DIMD mode may be derived based on restored samples belonging to a surrounding area of ​​the current block. In the video decoding method and device according to the present disclosure, the virtual intra prediction mode may be derived as a decoder side intra mode derivation (DIMD) mode. Here, the DIMD mode may be derived based on prediction samples belonging to a prediction block of the current block. In the video decoding method and device according to the present disclosure, the virtual intra prediction mode can be derived based on one or more TIMD (template-based intra mode derivation) modes for the current block. In the video decoding method and device according to the present disclosure, the virtual intra prediction mode may be derived based on at least one of whether an intra prediction mode corresponding to a division direction according to geometric division of the current block is identical to at least one of intra prediction modes applied to the plurality of partitions or a distance between a division line for geometric division of the current block and the center of the current block. The video encoding method and device according to the present disclosure can divide a current block into a plurality of partitions, perform prediction on each of the partitions to generate a prediction block of the current block, derive a residual block of the current block based on the prediction block of the current block, derive transform coefficients of the current block based on a transform for the residual block of the current block, and encode the transform coefficients of the current block. Here, the step of deriving the transform coefficients of the current block can include the step of deriving a virtual intra prediction mode for the current block, and the step of determining a transform set for the transform based on the virtual intra prediction mode. A computer-readable digital storage medium is provided, which stores encoded video / image information that causes a decoding device according to the present disclosure to perform a video decoding method. A computer-readable digital storage medium storing video / image information generated by a video encoding method according to the present disclosure is provided. A method and device for transmitting video / image information generated by a video encoding method according to the present disclosure are provided. According to the present disclosure, by using a non-separable primary transform as a primary transform, the performance of the transform can be improved. According to the present disclosure, the performance of the transformation can be improved by performing the transformation using a non-separable first-order transformation kernel of reduced dimension. According to the present disclosure, encoding efficiency can be improved by effectively determining and / or signaling a transform set and / or a transform kernel. FIG. 1 illustrates a video / image coding system according to the present disclosure. FIG. 2 is a schematic block diagram of an encoding device to which an embodiment of the present disclosure can be applied and in which encoding of a video / image signal is performed. FIG. 3 is a schematic block diagram of a decoding device to which an embodiment of the present disclosure can be applied and in which decoding of a video / image signal is performed. FIG. 4 illustrates an image decoding method performed by a decoding device (300) as an embodiment according to the present disclosure. FIG. 5 exemplarily shows an intra prediction mode and its prediction direction according to the present disclosure. Figure 6 illustrates an example of geometric segmentation according to the present disclosure. Fig. 7 is a flowchart illustrating a method for inducing a DIMD mode according to the present disclosure. Fig. 8 illustrates a filter for inducing a DIMD mode according to the present disclosure. FIG. 9 illustrates a schematic configuration of a decoding device (300) that performs an image decoding method according to the present disclosure. FIG. 10 illustrates an image encoding method performed by an encoding device (200) as an embodiment according to the present disclosure. FIG. 11 illustrates a schematic configuration of an encoding device (200) that performs an image encoding method according to the present disclosure. FIG. 12 illustrates an example of a content streaming system to which embodiments of the present disclosure can be applied. The present disclosure may be subject to various modifications and embodiments. Therefore, specific embodiments are illustrated and described in detail in the drawings. However, this is not intended to limit the present disclosure to specific embodiments, but rather to encompass all modifications, equivalents, and alternatives falling within the spirit and technical scope of the present disclosure. Similar reference numerals have been used to designate similar components throughout the description of each drawing. While terms such as "first" and "second" may be used to describe various components, these components should not be limited by these terms. These terms are used solely to distinguish one component from another. For example, without departing from the scope of the present disclosure, a first component could be referred to as a "second component," and similarly, a second component could also be referred to as a "first component." The term "and / or" includes a combination of multiple related items described herein or any of multiple related items described herein. When a component is referred to as being "connected" or "connected" to another component, it should be understood that it may be directly connected or connected to that other component, but that there may be other components intervening. Conversely, when a component is referred to as being "directly connected" or "connected" to another component, it should be understood that there are no other components intervening. The terminology used in this application is only used to describe specific embodiments and is not intended to limit the present disclosure. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, it should be understood that the terms "comprise" or "have" indicate the presence of a feature, number, step, operation, component, part, or combination thereof described in the specification, but do not preclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof. The present disclosure relates to video / image coding. For example, the methods / embodiments disclosed in this specification can be applied to methods disclosed in the versatile video coding (VVC) standard. In addition, the methods / embodiments disclosed in this specification can be applied to methods disclosed in the essential video coding (EVC) standard, the AOMedia Video 1 (AV1) standard, the second generation of audio video coding standard (AVS2), or the next generation of video / image coding standards (e.g., H.267 or H.268). This specification presents various embodiments of video / image coding, and unless otherwise stated, the embodiments may be performed in combination with each other. In this specification, a video may refer to a set of images over time. A picture generally refers to a unit representing one image at a specific time point, and a slice / tile is a unit that constitutes part of a picture in coding. A slice / tile may include one or more coding tree units (CTUs). A picture may be composed of one or more slices / tiles. A tile is a rectangular area consisting of multiple CTUs within a specific tile column and a specific tile row of a picture. A tile column is a rectangular area of ​​CTUs that has a height equal to the height of the picture and a width specified by the syntax requirements of the picture parameter set. A tile row is a rectangular area of ​​CTUs that has a height specified by the picture parameter set and a width equal to the width of the picture. CTUs within a tile are arranged consecutively according to the CTU raster scan, while tiles within a picture may be arranged consecutively according to the tile raster scan. A slice may contain an integer number of complete tiles or an integer number of contiguous complete CTU rows within a picture, which may be exclusively contained within a single NAL unit. Meanwhile, a picture may be divided into two or more subpictures. A subpicture may be a rectangular region of one or more slices within a picture. A pixel, or pel, can refer to the smallest unit that constitutes a picture (or image). Additionally, the term "sample" can be used as a counterpart to a pixel. A sample can generally represent a pixel or a pixel value, and can represent only the pixel / pixel value of the luminance component, or only the pixel / pixel value of the chrominance component. A unit may represent a basic unit of image processing. A unit may include at least one of a specific region of a picture and information related to the region. One unit may include one luma block and two chroma (e.g., cb, cr) blocks. In some cases, the term "unit" may be used interchangeably with terms such as "block" or "area." In general, an MxN block may include a set (or array) of samples (or sample array) or transform coefficients consisting of M columns and N rows. As used herein, "A or B" can mean "only A," "only B," or "both A and B." In other words, as used herein, "A or B" can be interpreted as "A and / or B." For example, as used herein, "A, B or C" can mean "only A," "only B," "only C," or "any combination of A, B and C." As used herein, a slash ( / ) or a comma can mean "and / or." For example, "A / B" can mean "A and / or B." Accordingly, "A / B" can mean "only A," "only B," or "both A and B." For example, "A, B, C" can mean "A, B, or C." In this specification, "at least one of A and B" may mean "only A", "only B" or "both A and B". Additionally, in this specification, the expressions "at least one of A or B" or "at least one of A and / or B" may be interpreted identically to "at least one of A and B". Additionally, in this specification, “at least one of A, B and C” can mean “only A,” “only B,” “only C,” or “any combination of A, B and C.” Additionally, “at least one of A, B or C” or “at least one of A, B and / or C” can mean “at least one of A, B and C.” Additionally, parentheses used herein may mean "for example." Specifically, when "prediction (intra-prediction)" is indicated, "intra-prediction" may be suggested as an example of "prediction." In other words, "prediction" in this specification is not limited to "intra-prediction," and "intra-prediction" may be suggested as an example of "prediction." Furthermore, even when "prediction (i.e., intra-prediction)" is indicated, "intra-prediction" may be suggested as an example of "prediction." Technical features individually described in a single drawing in this specification may be implemented individually or simultaneously. FIG. 1 illustrates a video / image coding system according to the present disclosure. Referring to FIG. 1, a video / image coding system may include a first device (source device) and a second device (receiving device). A source device can transmit encoded video / image information or data to a receiving device via a digital storage medium or a network in the form of a file or streaming. The source device may include a video source, an encoding device, and a transmitting device. The receiving device may include a receiving device, a decoding device, and a renderer. The encoding device may be referred to as a video / image encoding device, and the decoding device may be referred to as a video / image decoding device. The transmitter may be included in the encoding device. The receiver may be included in the decoding device. The renderer may include a display unit, and the display unit may be configured as a separate device or an external component. A video source may obtain video / images through a process of capturing, synthesizing, or generating video / images. The video source may include a video / image capture device and / or a video / image generation device. The video / image capture device may include one or more cameras, a video / image archive containing previously captured video / images, etc. The video / image generation device may include a computer, a tablet, a smartphone, etc., and may (electronically) generate video / images. For example, a virtual video / image may be generated through a computer, etc., in which case the video / image capture process may be replaced by a process of generating related data. An encoding device can encode input video / images. The encoding device can perform a series of procedures, such as prediction, transformation, and quantization, to improve compression and coding efficiency. The encoded data (encoded video / image information) can be output in the form of a bitstream. The transmission unit can transmit encoded video / image information or data output in the form of a bitstream to the receiving unit of a receiving device via a digital storage medium or network in the form of a file or streaming. The digital storage medium can include various storage media such as USB, SD, CD, DVD, Blu-ray, HDD, SSD, etc. The transmission unit can include an element for generating a media file via a predetermined file format and an element for transmission via a broadcasting / communication network. The receiving unit can receive / extract the bitstream and transmit it to a decoding device. The decoding device can decode the video / image by performing a series of procedures such as inverse quantization, inverse transformation, and prediction corresponding to the operation of the encoding device. The renderer can render decoded video / images. The rendered video / images can be displayed through the display unit. FIG. 2 is a schematic block diagram of an encoding device to which an embodiment of the present disclosure can be applied and in which encoding of a video / image signal is performed. Referring to FIG. 2, the encoding device (200) may be configured to include an image partitioner (210), a prediction unit (predictor) 220, a residual processor (residual processor) 230, an entropy encoder (entropy encoder) 240, an adder (adder) 250, a filter (filter) 260, and a memory (memory) 270. The prediction unit (220) may include an inter prediction unit (221) and an intra prediction unit (222). The residual processor (230) may include a transformer (transformer) 232, a quantizer (quantizer) 233, a dequantizer (dequantizer) 234, and an inverse transformer (inverse transformer) 235. The residual processing unit (230) may further include a subtractor (231). The addition unit (250) may be called a reconstructor or a recontructed block generator. The image segmentation unit (210), the prediction unit (220), the residual processing unit (230), the entropy encoding unit (240), the addition unit (250), and the filtering unit (260) described above may be configured by one or more hardware components (e.g., an encoding device chipset or processor) according to an embodiment. In addition, the memory (270) may include a decoded picture buffer (DPB) and may be configured by a digital storage medium. The hardware component may further include the memory (270) as an internal / external component. The image segmentation unit (210) can segment an input image (or picture, frame) input to the encoding device (200) into one or more processing units. For example, the processing unit may be called a coding unit (CU). In this case, the coding unit may be recursively segmented from a coding tree unit (CTU) or a largest coding unit (LCU) according to a QTBTTT (Quad-tree binary-tree ternary-tree) structure. For example, a single coding unit may be split into multiple coding units with deeper depths based on a quad-tree structure, a binary tree structure, and / or a ternary structure. In this case, for example, the quad-tree structure may be applied first, and the binary tree structure and / or the ternary structure may be applied later. Alternatively, the binary tree structure may be applied before the quad-tree structure. The coding procedure according to the present specification may be performed based on the final coding unit that is no longer split. In this case, based on coding efficiency according to image characteristics, etc., the largest coding unit may be used directly as the final coding unit, or, if necessary, the coding unit may be recursively split into coding units of lower depths, and the coding unit with the optimal size may be used as the final coding unit. Here, the coding procedure may include procedures such as prediction, transformation, and restoration, which will be described later. As another example, the processing unit may further include a prediction unit (PU) or a transform unit (TU). In this case, the prediction unit and the transform unit may each be split or partitioned from the final coding unit described above. The prediction unit may be a unit of sample prediction, and the transform unit may be a unit for deriving a transform coefficient and / or a unit for deriving a residual signal from a transform coefficient. The term "unit" may be used interchangeably with terms such as "block" or "area" depending on the case. In general, an MxN block can represent a set of samples or transform coefficients consisting of M columns and N rows. A sample can generally represent a pixel or a pixel value, and can represent only the pixel / pixel value of the luminance component, or only the pixel / pixel value of the chrominance component. A sample can be used as a term corresponding to a pixel or pel in a picture (or image). The encoding device (200) can generate a residual signal (residual block, residual sample array) by subtracting a prediction signal (prediction block, prediction sample array) output from an inter prediction unit (221) or an intra prediction unit (222) from an input video signal (original block, original sample array), and the generated residual signal is transmitted to a conversion unit (232). In this case, a unit that subtracts a prediction signal (prediction block, prediction sample array) from an input video signal (original block, original sample array) within the encoding device (200) may be called a subtraction unit (231). The prediction unit (220) can perform a prediction on a block to be processed (hereinafter, referred to as a current block) and generate a predicted block including prediction samples for the current block. The prediction unit (220) can determine whether intra prediction or inter prediction is applied on a current block or CU basis. The prediction unit (220) can generate various information related to prediction, such as prediction mode information, as described later in the description of each prediction mode, and transmit the information to the entropy encoding unit (240). The information related to prediction can be encoded by the entropy encoding unit (240) and output in the form of a bitstream. The intra prediction unit (222) can predict the current block by referring to samples in the current picture. The referenced samples may be located in the neighborhood of the current block, or may be located a certain distance away from the current block, depending on the prediction mode. In intra prediction, the prediction modes may include one or more non-directional modes and multiple directional modes. The non-directional mode may include at least one of a DC mode or a planar mode. The directional mode may include 33 directional modes or 65 directional modes depending on the degree of detail in the prediction direction. However, this is only an example, and a greater or lesser number of directional modes may be used depending on the settings. The intra prediction unit (222) may also determine the prediction mode applied to the current block by using the prediction mode applied to the neighboring blocks. The inter prediction unit (221) can derive a prediction block for the current block based on a reference block (reference sample array) specified by a motion vector on a reference picture. At this time, in order to reduce the amount of motion information transmitted in the inter prediction mode, the motion information can be predicted in units of blocks, subblocks, or samples based on the correlation of motion information between neighboring blocks and the current block. The motion information can include a motion vector and a reference picture index. The motion information can further include inter prediction direction information (L0 prediction, L1 prediction, Bi prediction, etc.). In the case of inter prediction, the neighboring block can include a spatial neighboring block existing in the current picture and a temporal neighboring block existing in the reference picture. The reference picture including the reference block and the reference picture including the temporal neighboring block may be the same or different. The above temporal neighboring blocks may be called collocated reference blocks, collocated CUs (colCUs), etc., and the reference pictures including the temporal neighboring blocks may be called collocated pictures (colPic). For example, the inter prediction unit (221) may construct a motion information candidate list based on the neighboring blocks, and generate information indicating which candidate is used to derive the motion vector and / or reference picture index of the current block. Inter prediction may be performed based on various prediction modes, and for example, in the case of skip mode and merge mode, the inter prediction unit (221) may use the motion information of the neighboring blocks as the motion information of the current block. In the case of skip mode, unlike the merge mode, a residual signal may not be transmitted.In the motion vector prediction (MVP) mode, the motion vector of the surrounding blocks is used as a motion vector predictor, and the motion vector of the current block can be indicated by signaling the motion vector difference. The prediction unit (220) can generate a prediction signal based on various prediction methods described below. For example, the prediction unit can apply intra prediction or inter prediction for prediction of a single block, and can also apply intra prediction and inter prediction simultaneously. This can be called combined inter and intra prediction (CIIP) mode. In addition, the prediction unit can be based on an intra block copy (IBC) prediction mode or a palette mode for prediction of a block. The IBC prediction mode or palette mode can be used for content image / video coding such as games, such as screen content coding (SCC). IBC basically performs prediction within the current picture, but can be performed similarly to inter prediction in that it derives a reference block within the current picture. That is, IBC can utilize at least one of the inter prediction techniques described herein. Palette mode can be viewed as an example of intra coding or intra prediction. When the palette mode is applied, sample values ​​within a picture can be signaled based on information about the palette table and palette index. The prediction signal generated through the prediction unit (220) can be used to generate a restoration signal or a residual signal. The transform unit (232) can apply a transform technique to the residual signal to generate transform coefficients. For example, the transform technique can include at least one of a Discrete Cosine Transform (DCT), a Discrete Sine Transform (DST), a Karhunen-Loeve Transform (KLT), a Graph-Based Transform (GBT), or a Conditionally Non-linear Transform (CNT). Here, GBT refers to a transform obtained from a graph when the relationship information between pixels is expressed as a graph. CNT refers to a transform obtained based on generating a prediction signal using all previously restored pixels. In addition, the transform process can be applied to a pixel block having a square size and the same size, or can be applied to a block of a non-square variable size. The quantization unit (233) quantizes the transform coefficients and transmits them to the entropy encoding unit (240), and the entropy encoding unit (240) can encode the quantized signal (information about the quantized transform coefficients) and output it as a bitstream. The information about the quantized transform coefficients can be called residual information. The quantization unit (233) can rearrange the quantized transform coefficients in a block form into a one-dimensional vector form based on the coefficient scan order, and can also generate information about the quantized transform coefficients based on the quantized transform coefficients in the one-dimensional vector form. The entropy encoding unit (240) can perform various encoding methods such as exponential Golomb, context-adaptive variable length coding (CAVLC), context-adaptive binary arithmetic coding (CABAC), etc. The entropy encoding unit (240) can also encode information necessary for video / image restoration (e.g., values ​​of syntax elements, etc.) together or separately from quantized transform coefficients. Encoded information (e.g., encoded video / image information) can be transmitted or stored in the form of a bitstream in units of NAL (network abstraction layer) units. The video / image information may further include information on various parameter sets, such as an adaptation parameter set (APS), a picture parameter set (PPS), a sequence parameter set (SPS), or a video parameter set (VPS). In addition, the video / image information may further include general constraint information. In the present specification, information and / or syntax elements transmitted / signaled from an encoding device to a decoding device may be included in the video / image information. The video / image information may be encoded through the above-described encoding procedure and included in the bitstream. The bitstream may be transmitted via a network or stored in a digital storage medium. Here, the network may include a broadcasting network and / or a communication network, and the digital storage medium may include various storage media, such as a USB, SD, CD, DVD, Blu-ray, HDD, or SSD. The signal output from the entropy encoding unit (240) may be configured as an internal / external element of the encoding device (200) by a transmitting unit (not shown) and / or a storing unit (not shown), or the transmitting unit may be included in the entropy encoding unit (240). The quantized transform coefficients output from the quantization unit (233) can be used to generate a prediction signal. For example, by applying inverse quantization and inverse transformation to the quantized transform coefficients through the inverse quantization unit (234) and the inverse transform unit (235), a residual signal (residual block or residual samples) can be restored. The addition unit (250) can generate a reconstructed signal (reconstructed picture, reconstructed block, reconstructed sample array) by adding the reconstructed residual signal to the prediction signal output from the inter prediction unit (221) or the intra prediction unit (222). When there is no residual for the block to be processed, such as when skip mode is applied, the predicted block can be used as a reconstructed block. The addition unit (250) may be called a restoration unit or a reconstructed block generation unit. The generated restoration signal can be used for intra prediction of the next processing target block within the current picture, and can also be used for inter prediction of the next picture after filtering as described below. Meanwhile, LMCS (luma mapping with chroma scaling) may be applied during the picture encoding and / or restoration process. The filtering unit (260) can improve subjective / objective picture quality by applying filtering to the restoration signal. For example, the filtering unit (260) can apply various filtering methods to the restoration picture to generate a modified restoration picture, and store the modified restoration picture in the memory (270), specifically, in the DPB of the memory (270). The various filtering methods can include deblocking filtering, sample adaptive offset, adaptive loop filter, bilateral filter, etc. The filtering unit (260) can generate various information regarding filtering and transmit it to the entropy encoding unit (240). The information regarding filtering can be encoded by the entropy encoding unit (240) and output in the form of a bitstream. The modified restored picture transmitted to the memory (270) can be used as a reference picture in the inter prediction unit (221). Through this, when inter prediction is applied, the encoding device can avoid prediction mismatch between the encoding device (200) and the decoding device, and can also improve encoding efficiency. The DPB of the memory (270) can store the modified restored picture to be used as a reference picture in the inter prediction unit (221). The memory (270) can store motion information of a block from which motion information is derived (or encoded) within the current picture and / or motion information of blocks within a picture that has already been restored. The stored motion information can be transferred to the inter prediction unit (221) to be used as motion information of a spatial neighboring block or motion information of a temporal neighboring block. The memory (270) can store restored samples of restored blocks within the current picture and transfer them to the intra prediction unit (222). FIG. 3 is a schematic block diagram of a decoding device to which an embodiment of the present disclosure can be applied and in which decoding of a video / image signal is performed. Referring to FIG. 3, the decoding device (300) may be configured to include an entropy decoder (310), a residual processor (320), a predictor (330), an adder (340), a filter (350), and a memory (360). The predictor (330) may include an inter-prediction unit (331) and an intra-prediction unit (332). The residual processor (320) may include a dequantizer (321) and an inverse transformer (321). The entropy decoding unit (310), residual processing unit (320), prediction unit (330), addition unit (340), and filtering unit (350) described above may be configured by a single hardware component (e.g., a decoding device chipset or processor) depending on the embodiment. In addition, the memory (360) may include a decoded picture buffer (DPB) and may be configured by a digital storage medium. The hardware component may further include the memory (360) as an internal / external component. When a bitstream including video / image information is input, the decoding device (300) can restore the image corresponding to the process in which the video / image information is processed in the encoding device of FIG. 2. For example, the decoding device (300) can derive units / blocks based on block division-related information obtained from the bitstream. The decoding device (300) can perform decoding using a processing unit applied in the encoding device. Accordingly, the processing unit of decoding may be a coding unit, and the coding unit may be divided from a coding tree unit or a maximum coding unit according to a quad tree structure, a binary tree structure, and / or a ternary tree structure. One or more transform units may be derived from the coding unit. Then, the restored image signal decoded and output through the decoding device (300) can be reproduced through a reproduction device. The decoding device (300) can receive a signal output from the encoding device of FIG. 2 in the form of a bitstream, and the received signal can be decoded through the entropy decoding unit (310). For example, the entropy decoding unit (310) can parse the bitstream to derive information (e.g., video / image information) necessary for image restoration (or picture restoration). The video / image information may further include information on various parameter sets, such as an adaptation parameter set (APS), a picture parameter set (PPS), a sequence parameter set (SPS), or a video parameter set (VPS). In addition, the video / image information may further include general constraint information. The decoding device can decode the picture further based on the information on the parameter set and / or the general constraint information. The signaling / received information and / or syntax elements described later in this specification can be decoded through the decoding procedure and obtained from the bitstream. For example, the entropy decoding unit (310) can decode information in a bitstream based on a coding method such as exponential Golomb coding, CAVLC, or CABAC, and output the values ​​of syntax elements required for image restoration and the quantized values ​​of transform coefficients for residuals. More specifically, the CABAC entropy decoding method receives a bin corresponding to each syntax element in the bitstream, determines a context model using information of the syntax element to be decoded and decoding information of the surrounding and decoding target blocks or information of symbols / bins decoded in the previous step, and predicts the occurrence probability of the bin according to the determined context model to perform arithmetic decoding of the bin to generate a symbol corresponding to the value of each syntax element.At this time, the CABAC entropy decoding method can update the context model using the information of the decoded symbol / bin for the context model of the next symbol / bin after determining the context model. Information regarding prediction among the information decoded by the entropy decoding unit (310) is provided to the prediction unit (inter prediction unit (332) and intra prediction unit (331)), and residual values ​​on which entropy decoding is performed by the entropy decoding unit (310), i.e., quantized transform coefficients and related parameter information, can be input to the residual processing unit (320). The residual processing unit (320) can derive a residual signal (residual block, residual samples, residual sample array). In addition, information regarding filtering among the information decoded by the entropy decoding unit (310) can be provided to the filtering unit (350). Meanwhile, a receiving unit (not shown) that receives a signal output from an encoding device may be further configured as an internal / external element of a decoding device (300), or the receiving unit may be a component of an entropy decoding unit (310). Meanwhile, a decoding device according to the present specification may be called a video / video / picture decoding device, and the decoding device may be divided into an information decoding device (video / video / picture information decoding device) and a sample decoding device (video / video / picture sample decoding device). The information decoding device may include the entropy decoding unit (310), and the sample decoding device may include at least one of the inverse quantization unit (321), the inverse transformation unit (322), the addition unit (340), the filtering unit (350), the memory (360), the inter prediction unit (332), and the intra prediction unit (331). The inverse quantization unit (321) can inverse quantize the quantized transform coefficients and output the transform coefficients. The inverse quantization unit (321) can rearrange the quantized transform coefficients into a two-dimensional block form. In this case, the rearrangement can be performed based on the coefficient scanning order performed in the encoding device. The inverse quantization unit (321) can perform inverse quantization on the quantized transform coefficients using quantization parameters (e.g., quantization step size information) and obtain transform coefficients. In the inverse transform unit (322), the transform coefficients are inversely transformed to obtain a residual signal (residual block, residual sample array). The prediction unit (320) can perform a prediction on the current block and generate a predicted block including prediction samples for the current block. The prediction unit (320) can determine whether intra-prediction or inter-prediction is applied to the current block based on the information regarding the prediction output from the entropy decoding unit (310), and can determine a specific intra / inter-prediction mode. The prediction unit (320) can generate a prediction signal based on various prediction methods described below. For example, the prediction unit (320) can apply intra prediction or inter prediction for prediction of a single block, and can also apply intra prediction and inter prediction simultaneously. This can be called combined inter and intra prediction (CIIP) mode. In addition, the prediction unit can be based on an intra block copy (IBC) prediction mode or a palette mode for prediction of a block. The IBC prediction mode or palette mode can be used for content image / video coding such as games, such as screen content coding (SCC). IBC basically performs prediction within the current picture, but can be performed similarly to inter prediction in that it derives a reference block within the current picture. That is, IBC can utilize at least one of the inter prediction techniques described herein. Palette mode can be viewed as an example of intra coding or intra prediction. When palette mode is applied, information about the palette table and palette index may be included and signaled in the video / image information. The intra prediction unit (331) can predict the current block by referring to samples within the current picture. The referenced samples may be located in the neighborhood of the current block, or may be located a certain distance away from the current block, depending on the prediction mode. In intra prediction, the prediction modes may include one or more non-directional modes and multiple directional modes. The intra prediction unit (331) may also determine the prediction mode applied to the current block by using the prediction mode applied to the neighboring blocks. The inter prediction unit (332) can derive a prediction block for the current block based on a reference block (reference sample array) specified by a motion vector on a reference picture. At this time, in order to reduce the amount of motion information transmitted in the inter prediction mode, the motion information can be predicted in units of blocks, subblocks, or samples based on the correlation of the motion information between the neighboring blocks and the current block. The motion information can include a motion vector and a reference picture index. The motion information can further include inter prediction direction information (L0 prediction, L1 prediction, Bi prediction, etc.). In the case of inter prediction, the neighboring blocks can include spatial neighboring blocks existing in the current picture and temporal neighboring blocks existing in the reference picture. For example, the inter prediction unit (332) can construct a motion information candidate list based on the neighboring blocks, and derive the motion vector and / or reference picture index of the current block based on the received candidate selection information. Inter prediction can be performed based on various prediction modes, and information about the prediction can include information indicating an inter prediction mode for the current block. The addition unit (340) can generate a restoration signal (restored picture, restoration block, restoration sample array) by adding the acquired residual signal to the prediction signal (prediction block, prediction sample array) output from the prediction unit (including the inter-prediction unit (332) and / or intra-prediction unit (331)). When there is no residual for the block to be processed, such as when skip mode is applied, the prediction block can be used as the restoration block. The addition unit (340) may be referred to as a restoration unit or restoration block generation unit. The generated restoration signal may be used for intra prediction of the next processing target block within the current picture, may be output after filtering as described below, or may be used for inter prediction of the next picture. Meanwhile, LMCS (luma mapping with chroma scaling) may be applied during the picture decoding process. The filtering unit (350) can improve subjective / objective image quality by applying filtering to the restored signal. For example, the filtering unit (350) can apply various filtering methods to the restored picture to generate a modified restored picture, and transmit the modified restored picture to the memory (360), specifically, to the DPB of the memory (360). The various filtering methods can include deblocking filtering, sample adaptive offset, adaptive loop filter, bilateral filter, etc. The (modified) reconstructed picture stored in the DPB of the memory (360) can be used as a reference picture in the inter prediction unit (332). The memory (360) can store motion information of a block from which motion information is derived (or decoded) in the current picture and / or motion information of blocks in a picture that has already been reconstructed. The stored motion information can be transferred to the inter prediction unit (260) to be used as motion information of a spatial neighboring block or motion information of a temporal neighboring block. The memory (360) can store reconstructed samples of reconstructed blocks in the current picture and transfer them to the intra prediction unit (331). In this specification, the embodiments described in the filtering unit (260), the inter prediction unit (221), and the intra prediction unit (222) of the encoding device (200) can be applied to the filtering unit (350), the inter prediction unit (332), and the intra prediction unit (331) of the decoding device (300) in the same or corresponding manner, respectively. FIG. 4 illustrates an image decoding method performed by a decoding device as an embodiment according to the present disclosure. Referring to FIG. 4, the current block can be divided into multiple partitions (S400). A current block can be divided into multiple partitions based on one or more dividing lines. The dividing lines can include at least one of a vertical line and a horizontal line. Alternatively, when geometric partitioning is applied to the current block, the current block can be divided into two partitions by a predetermined dividing line. The dividing line for the geometric partitioning can be defined based on a predetermined dividing direction (or dividing angle) and a distance from the center of the current block. The current block can be a coding block that is no longer divided through tree-based block partitioning. Referring to FIG. 4, a prediction block of the current block can be generated based on the prediction for each partition (S410). Assume that the current block is divided into two partitions, namely, a first partition and a second partition. In this case, the prediction block of the current block can be generated as a weighted sum of the first prediction block for the first partition and the second prediction block for the second partition. Here, the first and second prediction blocks can be generated based on intra prediction. Alternatively, the first and second prediction blocks can be generated based on inter prediction. Alternatively, either the first prediction block or the second prediction block can be generated based on intra prediction, and the other can be generated based on inter prediction. For example, if the current block is divided into two partitions based on geometric partitioning, a prediction block for each partition can be generated based on inter prediction, and a prediction block of the current block can be generated based on a weighted sum of the generated prediction blocks. Alternatively, if the current block is divided into two partitions based on geometric partitioning, a prediction block for each partition can be generated based on intra prediction, and the prediction block for the current block can be generated based on a weighted sum of the generated prediction blocks. This is hereinafter referred to as the Spatial Geometric Partitioning Mode (SGPM). When SGPM is applied to the current block, the partition type of the current block can be determined based on a partition type index that specifies the partitioning direction and position of the partition line. The partition type index can indicate any one of the predefined partition type candidates. The intra prediction mode of each partition within the current block can be derived based on a mode index that indicates any one of the multiple intra prediction mode candidates. The mode index can be defined for each partition within the current block. The above partition type index and the mode index for each partition can be signaled via the bitstream. In this case, the partition type index can be represented as partition_mode_idx, and the mode index for each partition can be represented as intra_pred_mode0_idx and intra_pred_mode1_idx, respectively. Alternatively, the partition type index and the mode index for the first partition can be signaled via the bitstream, and the mode index for the second partition can be derived based on the mode index for the first partition. At least one of the above partition type indexes or mode indexes can be signaled based on a flag (cu_sgpm_flag) indicating whether SGPM is applied to the current block. For example, when cu_sgpm_flag is 1, the partition type index and the mode index can be signaled, and when cu_sgpm_flag is 0, the partition type index and the mode index can not be signaled. Alternatively, a candidate list for SGPM may be constructed. The candidate list may include a plurality of candidates, each of which may include one partition type index and two mode indices. For example, the plurality of candidates in the candidate list may be derived from combinations of predefined partition type candidates (e.g., 26 partition type candidates) and predetermined intra prediction mode candidates (e.g., 3 intra prediction mode candidates). The maximum number of candidates that can be included in the candidate list may be 16. Based on any one of the plurality of candidates, a partition type index for the current block and a mode index for each partition may be derived. A candidate index indicating any one of the plurality of candidates may be signaled through a bitstream. The above candidate list may be reordered based on a predetermined template region. For example, a SAD between a predicted sample and a reconstructed sample in the template region may be calculated. The SAD may be calculated for each of a plurality of candidates in the candidate list. The plurality of candidates in the candidate list may be reordered in ascending order of the SAD. The template region may include at least one of a top peripheral region or a left peripheral region adjacent to the current block. The height of the top peripheral region and the width of the left peripheral region may be fixed to a predetermined length (e.g., 1). The above intra prediction mode candidates may be configured in an IPM list. An IPM list may be configured for each partition within the current block. At least one of the intra prediction mode candidates in the IPM list of the first partition may be different from the intra prediction mode candidates in the IPM list of the second partition. Alternatively, a single IPM list may be configured for the current block, and partitions within the current block may share the single IPM list. The IPM list may include three or more intra prediction mode candidates. SGPM can be applied when the size of the current block satisfies a certain condition. The condition may include at least one of the following conditions 1 to 5. Here, "width" and "height" may represent the width and height of the current block, respectively. (Condition 1) 4<=width<=64 (Condition 2) 4<=height<=64 (Condition 3) width <height*8 (Condition 4) height <width*8 (Condition 5) width*height>=32 A flag may be defined indicating whether blending between the prediction block of the first partition and the prediction block of the second partition for the current block is allowed. If the flag indicates that blending between prediction blocks is allowed (e.g., if the flag is false), adaptive blending may be used in the SGPM. The blending depth for the adaptive blending may be derived based on the size of the current block. For example, if the minimum of the width and height of the current block is 4, the blending depth may be derived as 1 / 2τ. If the minimum of the width and height of the current block is 8, the blending depth may be derived as τ. If the minimum of the width and height of the current block is 16, the blending depth may be derived as 2τ. If the minimum of the width and height of the current block is 32, the blending depth may be derived as 4τ. If the minimum of the width and height of the current block is greater than 32, the blending depth may be derived as 8τ. Here, τ may be any integer greater than 0. On the other hand, if the flag indicates that blending between prediction blocks is not allowed (e.g., if the flag is true), the blending depth can be derived as a default value (e.g., 1 / 4τ). This means that blending is not used when the dividing lines of the geometric segmentation correspond to vertical or horizontal lines, and the width of the region to which blending is applied becomes relatively narrower when the dividing lines of the geometric segmentation do not correspond to vertical or horizontal lines (i.e., when the dividing lines have different dividing directions). Referring to FIG. 4, a residual block of the current block can be derived based on the transform coefficients of the current block (S420). Residual information of a current block can be obtained from a bitstream, and the transform coefficients of the current block can be derived by decoding the residual information. A residual block including residual samples can be derived by performing at least one of dequantization and inverse transform on the transform coefficients of the current block. When an adaptive multiple transform selection (MTS) is applied to the current block, an inverse transform may be performed based on at least one of DCT-2, DST-7, or DCT-8. Here, DCT-2, DST-7, DCT-8, etc. may be referred to as a transform type, a transform kernel, or a transform core. In the present disclosure, the inverse transform may mean a separable transform. However, the present disclosure is not limited thereto, and the inverse transform may also mean a non-separable transform, or may be a concept that includes both a separable transform and a non-separable transform. In addition, the inverse transform in the present disclosure means a primary transform, but is not limited thereto, and may be transformed into an identical / similar form and applied to a secondary transform. For example, as a method for inverse transformation, only DCT-2 and a non-separable transform may be used, or a non-separable transform may be used in addition to at least one of DCT-2, DST-7, or DCT-8, or a non-separable transform may replace the transform kernel of one or more of DCT-2, DST-7, or DCT-8. As a more specific example, if there are (DCT-2, DCT-2), (DST-7, DST-7), (DCT-8, DST-7), (DST-7, DCT-8), (DCT-8, DCT-8) as transform kernel candidates for separable transform, a non-separable transform can replace or be added to one or more of the five transform kernel candidates. Here, the notation (transform1, transform2) indicates that transform1 is applied in the horizontal direction and transform2 is applied in the vertical direction. If a non-separable transform replaces some of the transform kernel candidates, the remaining transform kernel candidates except (DCT-2, DCT-2) and (DST-7, DST-7) can be replaced with the non-separable transform. However, the above transform kernel candidates are only examples, and other types of DCT and / or DST may be included, and a transform skip may be included as a transform kernel candidate. A non-separable transformation can refer to a transformation or inverse transformation based on a non-separable transformation matrix. That is, unlike a separable transformation that performs vertical and horizontal transformations independently by separating the vertical and horizontal transformations, a non-separable transformation can perform horizontal and vertical transformations simultaneously. For example, when a non-separable transformation is performed on a 4x4 block, the input data X to the non-separable transformation is as follows: When the above input data X is expressed in vector form, vector X' can be expressed as follows. In this case, the non-separable transformation can be performed as in the following mathematical expression 3. In mathematical expression 3, F represents a transformation coefficient vector, T represents a 16x16 non-separable transformation matrix, and ㆍ represents the multiplication of a matrix and a vector. A 16x1 transform coefficient vector F can be derived through the above mathematical expression 3, and the F can be reconstructed into 4x4 blocks according to a predetermined scan order. The scan order can be a horizontal scan, a vertical scan, a diagonal scan, a z-scan, a raster scan, or a predefined scan. The non-separable transform set and / or transform kernel for the above non-separable transform can be variously configured based on at least one of a prediction mode (e.g., intra mode, inter mode, etc.), the width, height, or number of pixels of the current block, the position of a sub-block within the current block, explicitly signaled syntax elements, statistical characteristics of surrounding samples, whether a secondary transform is used, or a quantization parameter (QP). Specifically, for the intra mode, pre-defined intra prediction modes are grouped to correspond to n sets of non-separable transformations, and each set of non-separable transformations may include k transform kernel candidates. Here, n and k may be arbitrary constants according to rules (conditions) defined identically for the encoding device and the decoding device. The number of non-separable transformation sets and / or the number of transformation kernel candidates included in the non-separable transformation sets may be configured differently depending on the width and / or height of the current block. For example, for a 4x4 block, n1 non-separable transformation sets and k1 transformation kernel candidates may be configured. For a 4x8 block, n2 non-separable transformation sets and k2 transformation kernel candidates may be configured. In addition, the number of non-separable transformation sets and the number of transformation kernel candidates included in each non-separable transformation set may be configured differently depending on the product of the width and height of the current block. For example, when the product of the width and height of the current block is equal to or greater than 256, n3 non-separable transformation sets and k3 transformation kernel candidates may be configured, and otherwise, n4 non-separable transformation sets and k4 transformation kernel candidates may be configured. That is, since the degree of change in the statistical characteristics of the residual signal varies depending on the block size, the number of non-separable transformation sets and transformation kernel candidates can be configured differently to reflect this. If the current block is divided into multiple sub-blocks, the statistical characteristics of the residual signal may be different for each sub-block, and therefore the number of non-separable transform sets and transform kernel candidates may be configured differently. For example, if a 4x8 or 8x4 block is divided into two 4x4 sub-blocks and a non-separable transform is applied to each sub-block, n5 non-separable transform sets and k5 transform kernel candidates may be configured for the upper left 4x4 sub-block, and n6 non-separable transform sets and k6 transform kernel candidates may be configured for the other 4x4 sub-blocks. Based on the explicitly signaled syntax element, the number of non-separable transformation sets and transformation kernel candidates can be configured differently. As the syntax element, information indicating any one of a plurality of non-separable transformation configurations can be used. For example, if three types of non-separable transformation configurations are supported (i.e., n7 non-separable transformation sets and k7 transformation kernel candidates, n8 non-separable transformation sets and k8 transformation kernel candidates, n9 non-separable transformation sets and k9 transformation kernel candidates), the corresponding syntax element can have values ​​of 0, 1, and 2, and the non-separable transformation configuration applied to the current block can be determined according to the value of the signaled syntax element. Depending on whether and / or which secondary transformation is applied, the number of non-separable transformation sets and transformation kernel candidates can be configured differently. For example, if no secondary transformation is applied, n 10 A set of non-separable transformations and k 10 A non-separable transformation configuration including n transformation kernel candidates can be applied. When a second transformation is applied, n 11 A set of non-separable transformations and k 11 A non-separable transformation configuration including a transformation kernel candidate can be applied. Depending on the quantization parameter (QP) and / or the range of QP values, different non-separable transform configurations can be applied. For example, if the QP value has a small value, n 12 A set of non-separable transformations and k 12 A non-separable transformation configuration including n transformation kernel candidates can be applied. On the other hand, if the QP value has a large value, n 13 A set of non-separable transformations and k 13A non-separable transform configuration including a candidate transform kernel can be applied. If the QP value is less than or equal to a threshold (e.g., 32), the case is classified as having a small QP value, and otherwise, the case is classified as having a large QP value. Alternatively, the QP value range can be divided into three or more ranges, and a different non-separable transform configuration can be applied to each range. For relatively large blocks, instead of using a non-separable transform corresponding to the width and height of the block, the block can be divided into multiple sub-blocks and a non-separable transform corresponding to the width and height of the sub-blocks can be used. For example, when performing a non-separable transform on a 4x8 block, the 4x8 block can be divided into two 4x4 sub-blocks and a 4x4 block-based non-separable transform can be used for each 4x4 sub-block. Alternatively, for an 8x16 block, the block can be divided into two 8x8 sub-blocks and an 8x8 block-based non-separable transform can be used. The above non-separable transform set can be determined based on the intra prediction mode of the current block and a mapping table. The mapping table can define a mapping relationship between pre-defined intra prediction modes and non-separable transform sets. The pre-defined intra prediction modes can include two non-directional modes and 65 directional modes. In general, the size of the transform kernel of a non-separable transform is larger than that of a separable transform. This means that the computational complexity required for the transform process is high and the memory required for storing the transform kernel is large. Meanwhile, a separable transform can only consider statistical characteristics existing in the horizontal and / or vertical directions, but a non-separable transform can simultaneously consider statistical characteristics in a two-dimensional space including the horizontal and vertical directions, thereby providing better compression efficiency. Since the statistical characteristics and diversity of the residual vary depending on the directionality of the intra prediction mode, there may be cases where a non-separable transform is absolutely necessary, and there may exist intra prediction modes where the characteristics of the residual can be sufficiently identified using only a separable transform. Therefore, by predefining which transformation to use according to the intra prediction mode in the encoding device and the decoding device, the transformation process can be designed with optimized complexity and memory requirements. The non-directional mode may include the planar mode (number 0) and the DC mode (number 1), and the directional mode may include the intra prediction modes (numbers 2 to 66). However, this is merely an example, and the present disclosure may also be applied to cases where the number of pre-defined intra prediction modes is different. Due to the application of wide angle intra prediction (WAIP), the pre-defined intra prediction modes may further include intra prediction modes from -14 to -1 and intra prediction modes from 67 to 80. FIG. 5 exemplarily illustrates intra prediction modes and their prediction directions according to the present disclosure. Referring to FIG. 5 , modes -14 to -1 and 2 to 33, and modes 35 to 80 are symmetrical with respect to the prediction direction with respect to mode 34. For example, modes 10 and 58 are symmetrical with respect to the direction corresponding to mode 34, and mode -1 is symmetrical with mode 67. Therefore, for vertical modes that are symmetrical with respect to horizontal modes with respect to mode 34, input data can be transposed and used. Transposing input data means that rows in the input data MxN of a 2D block become columns and columns become rows to form NxM data. For example, when a 4x4 block is used, the 16 data that make up the 4x4 block can be appropriately arranged to form a 16x1 one-dimensional vector for non-separable transformation. At this time, the one-dimensional vector can be formed in row-major order or column-major order. The residual samples resulting from the non-separable transformation can be arranged in the above order to form a two-dimensional block. For modes -14 to -1 and 2 to 33, if the data arrangement order for constructing a 16x1 input vector is row-major order, for modes 35 to 80, the input vector can be constructed according to column-major order. Although mode 34 can be considered neither a horizontal mode nor a vertical mode, it is classified as belonging to the horizontal mode in the present disclosure. That is, for modes -14 to -1 and 2 to 33, the input data alignment method for the horizontal mode, i.e., row-major order, can be used, and the input data can be transposed and used for the vertical mode that is symmetrical around mode 34. For non-square blocks, the symmetry in square blocks (i.e., the symmetry between the P mode and the (68-P) mode in an NxN block (2<=P<=33) or the symmetry between the Q mode and the (66-Q) mode (-14<=Q<=-1)) cannot be utilized. Therefore, in addition to the symmetry based only on the intra prediction mode, the symmetry between block shapes that are in a transpose relationship with each other, i.e., the symmetry between the KxL block and the LxK block, can also be utilized. Specifically, a symmetry relationship exists between a KxL block predicted by the P mode and an LxK block predicted by the (68-P) mode. Alternatively, a symmetry relationship exists between a KxL block predicted by the Q mode and an LxK block predicted by the (66-Q) mode. Since a KxL block with mode 2 and an LxK block with mode 66 can be viewed as symmetrical to each other, the same transform kernel can be applied to the KxL block and the LxK block. If a non-separable transform set for the intra prediction mode of the KxL block is mapped, in order to apply a non-separable transform to the LxK block, the non-separable transform set can be derived through a mapping table corresponding to the KxL block based on the (68-P) mode instead of the P mode applied to the LxK block. Alternatively, the non-separable transform set can be derived through a mapping table corresponding to the KxL block based on the (66-Q) mode instead of the Q mode applied to the LxK block. For example, to apply a non-separable transformation to an LxK block, the non-separable transformation set can be selected based on mode 2 instead of mode 66. Also, for a KxL block, the input data can be read in a pre-determined order (e.g., row-major order or column-major order) to form a one-dimensional vector and then the corresponding non-separable transformation can be applied. For an LxK block, the input data can be read in the transposed order to form a one-dimensional vector and then the corresponding non-separable transformation can be applied. That is, if the KxL block is read in row-major order, the LxK block can be read in column-major order. Conversely, if the KxL block is read in column-major order, the LxK block can be read in row-major order. In addition, when mode 34 is applied to a KxL block, a non-separable transformation set can be determined based on mode 34, and the input data can be read in a predetermined order to form a one-dimensional vector to perform the corresponding non-separable transformation. When mode 34 is applied to an LxK block, a non-separable transformation set can be determined based on mode 34 as well, but the input data can be read in a transposed order to form a one-dimensional vector to perform the corresponding non-separable transformation. Although the present disclosure describes a method for determining a non-separable transformation set and a method for organizing input data based on a KxL block, the non-separable transformation can be performed based on an LxK block by utilizing the symmetry described above for a KxL block in the same manner. Alternatively, a block whose width is greater than its height can be restricted to be used as a reference block. Alternatively, symmetry can be restricted not to be utilized for non-square blocks. In this case, non-square blocks can use a different number of non-separable transformation sets and / or transformation kernel candidates than square blocks, and can select a non-separable transformation set using a different mapping table than square blocks. An example of a mapping table for selecting a non-separable transformation set is as follows. predModeIntraTrSetIdxpredModeIntra < 040 <= predModeIntra <= 102 <= predModeIntra <= 12113 <= predModeIntra <= 23224 <= predModeIntra <= 44345 <= predModeIntra <= 55256 <= predModeIntra <= 66167 <= predModeIntra <= 804 Table 1 shows an example of assigning non-separable transform sets according to intra prediction modes when there are five non-separable transform sets. The value of predModeIntra indicates the value of the intra prediction mode considering WAIP, and TrSetIdx is an index indicating a specific non-separable transform set. In Table 1, it can be confirmed that the same non-separable transform set is applied to modes located in symmetrical directions according to the intra prediction mode. Table 1 is only an example using five non-separable transform sets and does not limit the total number of non-separable transform sets for non-separable transforms. Alternatively, as shown in Table 2, non-separable transform may not be applied to WAIP for compression performance. predModeIntraTrSetIdx0 <= predModeIntra <= 102 <= predModeIntra <= 12113 <= predModeIntra <= 23224 <= predModeIntra <= 44345 <= predModeIntra <= 55256 <= predModeIntra <= 661 Alternatively, as shown in Table 3, instead of constructing a separate non-separable transform set for WAIP, a non-separable transform set corresponding to an adjacent intra prediction mode may be shared. predModeIntraTrSetIdxpredModeIntra < 010 <= predModeIntra <= 102 <= predModeIntra <= 12113 <= predModeIntra <= 23224 <= predModeIntra <= 44345 <= predModeIntra <= 55256 <= predModeIntra <= 801 The above non-separable transform set may include multiple transform kernel candidates, and any one of the multiple transform kernel candidates may be selectively used. For this purpose, an index signaled through a bitstream may be used. Alternatively, any one of the multiple transform kernel candidates may be implicitly determined based on context information of the current block. Here, the context information may mean the size of the current block or whether a non-separable transform is applied to a neighboring block. Here, the size of the current block may be defined by the width, the height, the maximum / minimum values ​​of the width and the height, the sum of the width and the height, or the product of the width and the height. Below, we will take a closer look at how to determine the transformation kernel for the inverse transformation of the current block. Example 1 As mentioned above, inverse transformation can be divided into separable transformation and non-separable transformation. Separable transformation means performing transformations in the horizontal and vertical directions respectively for a two-dimensional block, and non-separable transformation can mean performing a single transformation on samples constituting the entire or a portion of a two-dimensional block. When notating a separable transformation, it can be notated as a pair of horizontal transformation and vertical transformation, and in this disclosure, it will be expressed as (horizontal transformation, vertical transformation). Multiple transformation sets can be defined for the inverse transformation of the current block. Each transformation set can contain one or more transformation kernel candidates. For example, any one of (DST-7, DST-7), (DCT-8, DST-7), (DST-7, DCT-8), or (DCT-8, DCT-8) can be applied as a separate transform, and the four transform kernel candidates can be regarded as one transform set. In addition, (DCT-2, DCT-2) can be regarded as one transform set. A transform skip that does not apply a transform can also be regarded as one transform set, and (DCT-2, DCT-2) and the transform skip can be regarded as one transform set. In the present disclosure, a transform kernel may refer to one transform (e.g., DCT-2, DST-7) or may refer to two transform pairs (e.g., (DCT-2, DCT-2)). As another example of a transform set, there may be the aforementioned non-separable transform set. In the present disclosure, the non-separable transform applied as the primary transform may be denoted as a Non-Separable Primary Transform (NSPT). In NSPT, multiple non-separable transform sets may be configured, and each non-separable transform set may include one or more transform kernels as transform kernel candidates. In the case of NSPT, one of the multiple non-separable transform sets is selected according to the intra prediction mode, and the multiple non-separable transform sets for NSPT may be denoted as an NSPT set list. This is as discussed above, and a detailed description thereof will be omitted here. A group of one or more transformation sets available to a current block can be formed from a plurality of pre-defined transformation sets. The group of one or more transformation sets can be formed by a predetermined area unit to which the current block belongs, and is hereinafter referred to as a collection. Here, the predetermined area unit can be at least one of a picture, a slice, a coding tree unit row (CTU row), or a coding tree unit (CTU). For example, let the transform set consisting of (DCT-2, DCT-2) be called S1, and the transform set consisting of (DST-7, DST-7), (DCT-8, DST-7), (DST-7, DCT-8) and (DCT-8, DCT-8) be called S2, respectively. In addition, the above-mentioned NSPT set list can include N non-separable transform sets, and the N non-separable transform sets are called S 3,1 , S 3,2 , ..., S 3,N Let us call them respectively. Here, N can be 35, but is not limited thereto. S by intra prediction mode of current block 3,13 When selected as a non-separable transformation set for this NSPT, the transformation kernels applicable to the current block are S1, S2, or S 3,13 It may belong to any of the following collections. In this case, the current block is available in {S1, S2, S 3,13} can be expressed as . As described above, since a collection according to the present disclosure is a group of one or more transformation sets available to the current block, the collection can be configured differently depending on the context of the current block. Here, the context can include at least one of a shape, a size, or an intra prediction mode. If a total of K contexts are defined, K collections can be created, each collection having C i can be represented as (i=1, 2, ..., N). For example, if the block sizes to which NSPT can be applied are 4x4, 8x8, 16x16, 32x32, and one of a total of 35 non-separable transformation sets is selected by the intra prediction mode, if the transformation kernel applied to each block size is different, a total of 4 x 35 = 140 contexts can be defined. A collection may be constructed based on the context of the current block, and at this time, a process of selecting one of a plurality of transformation sets belonging to the collection and selecting one of a plurality of transformation kernel candidates belonging to the selected transformation set may be performed. Here, the selection of the transformation set and the transformation kernel candidate may be performed implicitly based on the context of the current block or based on an explicitly signaled index. Alternatively, the process of selecting one of a plurality of transformation sets belonging to the collection and the process of selecting one of a plurality of transformation kernel candidates belonging to the selected transformation set may be performed separately. For example, an index for selecting a transformation set may be first signaled, and based on this, one of the plurality of transformation sets belonging to the collection may be selected. Then, an index indicating one of a plurality of transformation kernel candidates belonging to the transformation set may be signaled, and based on the signaled index, one of the transformation kernel candidates may be selected from the transformation set. The transformation kernel of the current block may be determined based on the selected transformation kernel candidate. Alternatively, the selection of any one of the transformation sets from the collection may be performed implicitly based on the context of the current block, and the selection of any one of the transformation kernel candidates from the selected transformation set may be performed based on a signaled index. Alternatively, the selection of any one of the transformation sets from the collection may be performed based on a signaled index, and the selection of any one of the transformation kernel candidates from the selected transformation set may be performed implicitly based on the context of the current block. Alternatively, the selection of any one of the transformation sets from the collection may be performed implicitly based on the context of the current block, and the selection of any one of the transformation kernel candidates from the selected transformation set may also be performed implicitly based on the context of the current block. Of course, if the number of transformation sets belonging to the collection is 1, the index for selecting the transformation set may not be signaled.Similarly, if the number of transformation kernel candidates in the selected transformation set is only one, the index indicating the transformation kernel candidate may not be signaled. Alternatively, an index indicating any one of all transformation kernel candidates in the current collection may be signaled. In this case, the process of selecting a transformation set from the collection may be omitted. At this time, all transformation sets in the collection may be reordered considering priorities. For example, in cases where small-length binary codes are assigned to small-value indices, such as truncated unary codes, it may be advantageous to assign small-value indices to transformation kernel candidates that are more advantageous for improving coding performance. When reordering all transformation kernel candidates in a collection according to priorities (shuffling), different shuffling may be applied to each collection. Furthermore, rather than reordering all transformation kernel candidates in a collection, only some of them may be selectively reordered. Example 2 The transformation kernel for the inverse transformation of the current block can be determined based on MTS (Multiple Transform Selection). The MTS according to the present disclosure may use at least one of DST-7, DCT-8, DCT-5, DST-4, DST-1, or IDT (identity transform) as a transform kernel. In addition, the MTS according to the present disclosure may further include a DCT-2 transform kernel. In the present disclosure, multiple MTS sets for MTS can be defined. Based on the size of the current block and / or the intra prediction mode, any one of the multiple MTS sets can be determined. For example, in determining any one MTS set, 16 transform block sizes can be considered, and for directional modes, the shape of the transform block and the symmetry between the intra prediction modes can be considered. For the WAIP (Wide Angle Intra Prediction) mode (i.e., -1 to -14 (or -15), 67 to 80 (or 81)), the MTS set corresponding to mode 2 can be applied to modes -1 to -14 (or -15), and the MTS set corresponding to mode 66 can be applied to modes 67 to 80 (or 81). A separate MTS set can be allocated for the MIP (Matrix-based Intra Prediction) mode. For example, the MTS set according to the transform block size and intra prediction mode can be allocated / defined as shown in Table 4 below. Block size Intra prediction mode Width Height [0, 1] [2, 12] [13, 23] [24, 34] MIP4 40 1 2 3 4 4 8 5 6 7 8 9 4 16 10 1 1 2 1 3 1 4 4 3 2 1 5 1 6 1 7 1 8 1 9 8 4 2 0 2 1 2 2 2 3 2 4 8 8 2 5 2 6 2 7 2 8 2 9 8 16 3 0 3 1 3 2 3 3 3 4 8 3 2 3 5 3 6 3 7 3 8 3 9 16 4 4 0 4 1 4 2 4 3 4 1 6 8 4 5 4 6 4 7 4 8 4 9 16 1 6 5 0 5 1 5 2 5 3 5 4 1 6 3 2 5 5 5 6 5 7 5 8 5 9 3 2 4 6 0 6 16 2 6 3 6 4 3 2 8 6 5 6 6 6 7 6 8 6 9 3 2 1 6 7 0 7 1 7 2 7 3 7 4 3 2 3 2 7 5 7 6 7 7 7 8 7 9 Table 4 shows the allocation of MTS sets according to 16 transform block sizes and intra prediction modes. The number of predefined MTS sets is 80, and the index indicating any one of the 80 MTS sets can range from 0 to 79, as shown in Table 4. MTS set index transformation kernel candidate index Table 5 shows the transformation kernel candidates included in each MTS set discussed in Table 4. Each MTS set can be composed of six transformation kernel candidates. The transformation kernel candidate index has a value of any one of 0 to 5 and can indicate any one of the six transformation kernel candidates. Here, each transformation kernel candidate can be a combination of a horizontal transformation kernel and a vertical transformation kernel for a separate transformation, and 25 transformation kernel candidates with indices of 0 to 24 can be defined. Kernel Combination Index When the value of intra prediction mode is less than 35 When the value of intra prediction mode is 35 or more 0 (DCT-8, DCT-8) (DCT-8, DCT-8) 1 (DST-7, DCT-8) (DCT-8, DST-7) 2 (DCT-5, DCT-8) (DCT-8, DCT-5) 3 (DST-4, DCT-8) (DCT-8, DST-4) 4 (DST-1, DCT-8) (DCT-8, DST-1) 5 (DCT-8, DST-7) (DST-7, DCT-8) 6 (DST-7, DST-7) (DST-7, DST-7) 7 (DCT-5, DST-7) (DST-7, DCT-5) 8 (DST-4, DST-7) (DST-7, DST-4) 9 (DST-1, DST-7)(DST-7, DST-1)10(DCT-8, DCT-5)(DCT-5, DCT-8)11(DST-7, DCT-5)(DCT-5, DST-7)12(DCT-5, DCT-5)(DCT-5, DCT-5)13(DST-4, DCT-5)(DCT-5, DST-4)14(DST-1, DCT-5)(DCT-5, DST-1)15(DCT-8, DST-4)(DST-4, DCT-8)16(DST-7, DST-4)(DST-4, DST-7)17(DCT-5, DST-4)(DST-4, DCT-5)18(DST-4, DST-4)(DST-4, DST-4)19(DST-1, DST-4)(DST-4, DST-1)20(DCT-8, DST-1)(DST-1, DCT-8)21(DST-7, DST-1)(DST-1, DST-7)22(DCT-5, DST-1)(DST-1, DCT-5)23(DST-4, DST-1)(DST-1, DST-4)24(DST-1, DST-1)(DST-1, DST-1) Table 6 is an example of the 25 transform kernel candidates discussed in Table 5. Specifically, the horizontal and vertical transforms of the transform kernel candidates are indicated as (horizontal transform, vertical transform). For each transform kernel candidate index, the horizontal / vertical transform when the intra prediction mode is less than 35 may be the opposite of the horizontal / vertical transform when the intra prediction mode is 35 or greater. When the value of the intra prediction mode is 35 or greater, a mode symmetrical around mode 34 may be derived, and an MTS set may be selected from Table 4 based on the mode. In addition, the symmetry of the block shape may be additionally considered. When the original transform block has a WxH size, the original transform block may be considered to have a HxW size by symmetrizing it, and an MTS set may be selected from Table 4. Here, the value of the intra prediction mode may be the value of the modified intra prediction mode. That is, as mode values ​​for WAIP, -14 (or -15) to -1 are modified to mode 2, 67 to 80 (or 81) are modified to mode 66, and the remaining modes can be set to the values ​​of the modified intra prediction modes with the values ​​of the original intra prediction modes as they are. In this case, since the extended modes for WAIP are also configured symmetrically around mode 34, the symmetry around mode 34 can be utilized for all directional modes except for the Planar mode and the DC mode. For example, if a 16x32 block is predicted to be mode 54, mode 14 (=68-54) can be derived as a mode symmetric to mode 54, and the block size can be considered as 32x16. In this case, an MTS set with an index of 72 can be selected, as defined in Table 4. When the MIP mode is applied, the MTS set assigned to the MIP mode may be selected based on the size of the current block, without considering the symmetry of the block shape. Alternatively, when the MIP mode is applied, the MTS set assigned to the MIP mode may be selected based on the symmetric block size, considering the symmetry of the block shape. For example, when the MIP mode is applied to an 8x16 block, the 8x16 block may be regarded as a symmetrical 16x8 block, and an MTS set with an index of 49 may be selected, as defined in Table 4. Alternatively, when the MIP mode is applied, the intra prediction mode may be regarded as the Planar mode. In this case, the MTS set assigned to the MIP mode may be selected based on the size of the current block, without considering the symmetry of the block shape. Alternatively, the MTS set assigned to the MIP mode may be selected based on the symmetrical block size, considering the symmetry of the block shape. For the MIP mode, a flag may be used to indicate whether the MIP mode is applied in transpose mode. If the MIP mode is applied to the current block of MxN and the flag indicates application of the transpose mode, the intra prediction mode is regarded as the Planar mode, and the current block of MxN may be regarded as an NxM block. That is, from Table 4, an MTS set corresponding to the block size of NxM and the Planar mode may be selected. As seen in Table 6, if the value of the intra prediction mode is 35 or greater, the horizontal and vertical transformations are swapped, but since the intra prediction mode of the current block is regarded as the Planar mode, the horizontal and vertical transformations of the transform kernel candidates may not be swapped. Alternatively, if the MIP mode is applied to the current block of MxN and the flag indicates application of the transpose mode, the intra prediction mode is not regarded as the Planar mode, and the current block of MxN may be regarded as an NxM block. That is, from Table 4, an MTS set corresponding to the block size of NxM and the MIP mode may be selected. In Table 5, a transformation kernel candidate selected by a transformation kernel candidate index may be set as the transformation kernel of the current block. Alternatively, depending on the size of the current block, at least one of the horizontal transformation or the vertical transformation of the selected transformation kernel candidate may be changed to another transformation kernel. For example, if the transformation kernel candidate index is 3 and the width and height of the current block are both 16 or less, at least one of the horizontal transformation or the vertical transformation of the transformation kernel candidate corresponding to the transformation kernel candidate index of 3 may be changed to another transformation kernel. At this time, the horizontal transformation and the vertical transformation may be changed independently of each other. If the difference (or the absolute value of the difference) between the value of the intra prediction mode of the current block and the value of the horizontal mode is less than or equal to a predetermined threshold, the vertical transformation of the selected transformation kernel candidate may be changed to an identity transform (IDT). If the difference (or the absolute value of the difference) between the value of the intra prediction mode of the current block and the value of the vertical mode is less than or equal to a predetermined threshold, the horizontal transformation of the selected transformation kernel candidate may be changed to an identity transform (IDT). Here, the threshold value can be determined based on the width and height of the current block, as shown in Table 7 below. Block Size Threshold Width Height 44848641648488888166164416821616-1 Table 7 defines thresholds according to the size of a transform block for changing the horizontal transformation and / or vertical transformation of a transform kernel candidate selected by a transform kernel candidate index to another transform kernel. The six transform kernel candidates constituting one MTS set can be distinguished by transform kernel candidate indices from 0 to 5 as defined in Table 5. The transform kernel candidate indices can be signaled through the bitstream. A flag indicating whether the MTS set is available / applied (MTS enabled flag or MTS flag) can be signaled, and the transform kernel candidate index can be signaled when the flag indicates the availability / applicability of the MTS set. The MTS flag can be composed of one bin, and one or more contexts for CABAC-based entropy coding (hereinafter referred to as CABAC context) can be allocated to the bin. For example, different CABAC contexts can be allocated for non-MIP mode and MIP mode, respectively. Depending on the context of the current block described above, the number of transform kernel candidates available to the current block may be set differently. For example, as the context of the current block, the sum of the absolute values ​​of all or part of the transform coefficients in the current block may be considered. The sum of the absolute values ​​of the transform coefficients is referred to as AbsSum. If AbsSum is less than or equal to T1, only one transform kernel candidate corresponding to a transform kernel candidate index of 0 may be available. If AbsSum is greater than T1 and less than or equal to T2, four transform kernel candidates corresponding to transform kernel candidate indices of 0 to 3 may be available. If AbsSum is greater than T2, six transform kernel candidates corresponding to transform kernel candidate indices of 0 to 5 may be available. Here, T1 may be 6 and T2 may be 32, but this is only an example. When AbsSum is less than or equal to T1, since the number of transformation kernel candidates available for the current block is 1, the transformation kernel candidate corresponding to the transformation kernel candidate index of 0 can be set as the transformation kernel of the current block without signaling the transformation kernel candidate index. When AbsSum is greater than T1 and less than or equal to T2, since four transformation kernel candidates are available, any one of the four transformation kernel candidates can be selected based on the transformation kernel candidate index with two bins. That is, the transformation kernel candidate indices of 0 to 3 can be signaled as 00, 01, 10, and 11, respectively. For the two bins, the Most Significant Bit (MSB) can be signaled first, and the Least Significant Bit (LSB) can be signaled later. Different CABAC contexts can be assigned to each bin. For example, for two bins, a CABAC context other than the CABAC context assigned for the MTS flag may be assigned to each bin. Alternatively, bypass coding may be applied without assigning a CABAC context to the two bins. When AbsSum is greater than T2, the transform kernel candidate index has a value from 0 to 5, so the transform kernel candidate index cannot be expressed with only two bins. In this case, the transform kernel candidate index may be expressed by assigning two or more bins, such as truncated binary coding. For each bin assigned by the truncated binary coding method, a CABAC context may be assigned, or bypass coding may be applied without assigning a CABAC context. Alternatively, a CABAC context may be assigned to some of the multiple bins (e.g., the first bin, or the first and second bins), and bypass coding may be applied to the remaining bins. Example 3 The transformation kernel of the current block can be determined based on a transformation set including one or more transformation kernel candidates. The transformation kernel of the current block can be derived from any one or more transformation kernel candidates belonging to the transformation set. The process of determining a transformation kernel of the current block may include at least one of 1) determining a transformation set of the current block or 2) selecting one transformation kernel candidate from the transformation set of the current block. The process of determining the transformation set may be a process of selecting one of a plurality of transformation sets that are identically predefined for the encoding device and the decoding device. Alternatively, the process of determining the transformation set may be a process of configuring one or more transformation sets available to the current block from among a plurality of transformation sets that are identically predefined for the encoding device and the decoding device, and selecting one of the configured transformation sets. Alternatively, the process of determining the transformation set may be a process of configuring one transformation set based on a transformation kernel candidate available to the current block from among a plurality of transformation kernel candidates that are identically predefined for the encoding device and the decoding device. If the transformation set of the current block includes multiple transformation kernel candidates, a process of selecting one of the multiple transformation kernel candidates for the current block may be performed. However, if the transformation set of the current block includes only one transformation kernel candidate (i.e., the current block has only one transformation kernel candidate available), the transformation kernel of the current block may be set to that transformation kernel candidate. The transformation set according to the present disclosure may mean the (non-separable) transformation set in the aforementioned embodiment 1, or may mean the MTS set in the embodiment 2. Alternatively, the transformation set may be defined separately from the (non-separable) transformation set in the embodiment 1 or the MTS set in the embodiment 2. In this case, the transformation set may include one or more specific transformation kernels as transformation kernel candidates. One specific transformation kernel may be defined as a pair of a transformation kernel for horizontal transformation and a transformation kernel for vertical transformation, or may be defined as one transformation kernel that is equally applied to horizontal and vertical transformations. In the embodiments of the present disclosure, the process of applying NSPT, which is a non-separable transform applied as a primary transform, is described in detail. NSPT can be applied to all or part of a transform block. Based on the forward NSPT, residual samples existing in the region where NSPT is applied can be input as a one-dimensional vector of NSPT. That is, residual samples existing in all or part of a transform block (referred to as Region Of Interest, ROI, in the present disclosure) can be collected into a one-dimensional vector and configured as an input. Then, by applying the forward NSPT, the primary transform coefficients can be obtained. Conversely, by applying the backward NSPT to the primary transform coefficients, the one-dimensional vector output can be obtained. By arranging each element value constituting the corresponding output vector at a predetermined location within the 2D transform block, the residual sample for the ROI can be obtained. The dimension of the matrix of the non-separable transform kernel for NSPT may be determined according to the size of the ROI. In the present disclosure, the transform kernel may be referred to as a transform type and a transform matrix, and the non-separable transform kernel for NSPT may be referred to as an NSPT kernel. For example, if the current block is an MxN transform block, the ROI is the area of ​​the entire MxN transform block, and a square NSPT is applied, the dimension of the corresponding transform matrix may be MN x MN. For example, if the ROI is the area of ​​the entire 8x8 transform block, the dimension of the NSPT kernel may be 64 x 64. According to one embodiment of the present disclosure, when applying NSPT to a residual generated by intra prediction, the NSPT kernel can be adaptively determined depending on the intra prediction mode. Since the statistical characteristics of the residual block may vary depending on the intra prediction mode, compression efficiency can be improved by adaptively determining the NSPT kernel depending on the intra prediction mode. It can be configured to share an NSPT kernel applied to one or more intra prediction modes. As described above, the non-separable transform set can be determined based on the intra prediction mode of the current block and a mapping table. The mapping table can define a mapping relationship between pre-defined intra prediction modes and non-separable transform sets. The pre-defined intra prediction modes can include two non-directional modes and 65 directional modes. In one embodiment, intra prediction modes can be grouped into intra prediction mode groups. One NSPT kernel or multiple NSPT kernels can be assigned to an intra prediction mode group. In other words, a non-separable transform set (NSPT set) including one or more NSPT kernels can be assigned to an intra prediction mode group. The non-separable transform set is mapped to an intra prediction mode, and one of the N NSPT kernels included in the non-separable transform set can be selected. For example, an intra prediction group may include adjacent prediction modes (e.g., modes 17, 18, and 19). In addition, the intra prediction group may include modes that have symmetry. For example, directional modes may be symmetrical around the diagonal mode of FIG. 5 described above (i.e., intra prediction mode 34). In this case, two modes that have symmetry may form a group (or pair). For example, modes 18 and 50 may be included in the same group because they are symmetrical around mode 34. However, for modes that have symmetry, a process of transposing the 2D input block and then constructing a one-dimensional input vector may be added before applying the forward NSPT kernel. For example, if the intra prediction mode is 34 or less, the one-dimensional input vector may be derived from the corresponding input block in the row-first order without transposing the 2D input block. If the intra prediction mode is greater than 34, the one-dimensional input vector can be constructed by first transposing the 2D input block and then reading the input block in the row-first order, or by leaving the 2D input block as is and reading the input block in the column-first order. Table 8 below illustrates an example of a mapping table for assigning NSPT sets according to intra prediction modes. Referring to Table 8, a total of 35 NSPT sets, from 0 to 34, can be defined. The extended WAIP mode (i.e., modes -14 to -1 and modes 67 to 80 in FIG. 5 ) can be assigned the NSPT set assigned to the nearest normal directional mode. In other words, the extended WAIP mode can be assigned NSPT set number 2. An NSPT set may include one or more NSPT kernels (or kernel candidates). That is, an NSPT set may include N NSPT kernel candidates. For example, N may be set to a value greater than or equal to 1, such as 1, 2, 3, or 4. Among one or more NSPT kernels included in the NSPT set, a kernel applied to a current block may be signaled using an index. In the present disclosure, the index may be referred to as an NSPT index. For example, the NSPT index may have values ​​of 0, 1, 2, …, N - 1. Additionally, as an example, if the number of NSPT kernel candidates is 1, the NSPT index value may be fixed to 0. In this case, the NSPT index may be inferred without being separately signaled. In addition, a flag indicating whether NSPT is applied may be signaled separately from the NSPT index. In the present disclosure, the flag may be referred to as an NSPT flag. If the NSPT flag value is 1, NSPT may be applied. If the NSPT flag value is 0, NSPT may not be applied. If the NSPT flag is not signaled, the NSPT flag value may be inferred to be 0. For example, if the NSPT flag value is 1, the NSPT index may be signaled. Based on the signaled NSPT index, one of the N kernel candidates included in the NSPT set selected by the intra prediction mode may be specified. In one embodiment, the entropy coding method of the NSPT index can be defined in various ways considering the number (N) of NSPT kernels included in the NSPT set. For example, as a method of mapping values ​​from 0 to N-1 to empty strings (i.e., a binarization method), truncated unary binarization, truncated binarization, and fixed-length binarization methods can be used. For example, if the number of kernel candidates constituting the NSPT set, N, is 2, one of the two candidates can be specified by one bin. For example, 0 can indicate the first candidate, and 1 can indicate the second candidate. In addition, if the value of N is 3 and truncated unary binarization is applied, the candidates can be specified by two bins. For example, the first, second, and third candidates can be binarized and signaled as 0, 10, and 11, respectively. As an example, the binarized bins can be coded using context coding or bypass coding. In this disclosure, a reduced primary transform (RPT) method using a transform kernel of a reduced dimension by a primary transform is described. As described above, when forward NSPT is applied, samples belonging to a 2D residual block can be arranged (or rearranged) into a 1D vector according to row-majority (or column-majority). Thereafter, a transformation matrix for NSPT can be multiplied by the arranged vector. When the corresponding 2D residual block is an M x N block (M is the width, N is the height), the length of the rearranged 1D vector can be M*N. That is, the corresponding 2D residual block can also be represented as a column vector having dimensions M*N x 1. In this disclosure, M*N can be conveniently denoted as MN. In this case, the dimension of the corresponding transformation matrix can be MN x MN. In summary, forward NSPT can operate by multiplying the left side of an MN x 1 vector by the corresponding MN x MN transformation matrix to obtain an MN x 1 transformation coefficient vector. When RPT is applied, instead of multiplying the MN x MN matrix as the forward NSPT transformation matrix described above, r transform coefficients can be obtained by multiplying an r x MN matrix. Here, r represents the number of rows of the transformation matrix, and MN represents the number of columns of the transformation matrix. According to an embodiment of the present disclosure, the value of r can be set to be less than or equal to MN. That is, the existing forward NSPT transformation matrix includes MN rows, each row being a 1 x MN row vector, which is a transform basis vector of the corresponding NSPT transformation matrix. The corresponding transform coefficients can be obtained by multiplying each transform basis vector by an MN x 1 sample column vector. Since the conventional forward NSPT transformation matrix is ​​composed of MN row vectors, MN transformation coefficients (i.e., MN x 1 transformation coefficient column vectors) can be obtained by applying the forward NSPT. On the other hand, in the case of the forward RPT, the transformation matrix can be composed of r transformation basis vectors instead of MN transformation basis vectors. Accordingly, when the forward RPT is applied, r transformation coefficients (i.e., r x 1 transformation coefficient column vectors) can be obtained instead of MN. The RPT kernel can be constructed by selecting r transform basis vectors, which are part of the transform basis vectors constituting the MN x MN forward NSPT kernel. In the present disclosure, the transform kernel can be referred to as a transform type and a transform matrix, and the non-separable transform kernel for NSPT can be referred to as an RPT kernel. That is, when selecting r 1 x MN row vectors from the MN x MN forward NSPT kernel, it can be advantageous to select the most important transform basis vectors from the perspective of coding performance. Specifically, in terms of energy compaction through transform, more energy can be concentrated in the transform coefficients that appear first by multiplying the forward NSPT transform matrix. In other words, the transform basis vectors located on the upper side of the forward NSPT transform matrix can generate transform coefficients having larger energy. Considering this, the r x MN forward RPT kernel can be constructed (or derived) by taking r from the upper side of the forward NSPT kernel. The RPT according to the present disclosure takes only a portion (i.e., r) of the transform coefficients obtained by applying the existing NSPT, which may result in a loss of some of the energy contained in the original signal. In other words, distortion may occur between the original signal and the original signal through this process. Nevertheless, by applying the RPT, only r transform coefficients are generated instead of MN, so the number of bits required to code the transform coefficients can be reduced. Therefore, in the case of a signal in which a large amount of energy is concentrated in a small number of transform coefficients (e.g., an image residual signal), the gain obtained by reducing the signaling bits can be significantly large, thereby improving the coding performance. The backward NSPT may be the transpose matrix of the forward NSPT kernel described above as a transformation matrix. In this case, the input data may be a transformation coefficient signal instead of a sample signal such as a residual signal. Specifically, if the forward NSPT transformation matrix is ​​G and the sample signal rearranged into a 1D vector is x, the transformation coefficient vector obtained by multiplying the transformation matrix on the left side can be expressed as in the following mathematical equation 4. Referring to Equation 4, x and y can be MN x 1 column vectors. G can have the form of an MN x MN matrix. The reverse NSPT process can be expressed as Equation 5 below using the same variables. In mathematical expression 5, G T means the transpose matrix of G. The forward RPT operation and the backward RPT operation according to the present disclosure can also be expressed by the two mathematical equations above. However, when RPT is applied, y is an r x 1 column vector instead of an MN x 1 column vector, and G is an r x MN matrix instead of an MM x MN matrix. That is, even if RPT is applied instead of NSPT, the dimension of the sample signal (e.g., image residual signal) does not change, which may mean that the original number of sample signals (i.e., MN sample signals) can be restored with only r transform coefficients through the backward RPT. In other words, the original MN sample signals can be restored by coding only r transform coefficients that are less than MN, which may lead to an improvement in coding performance. In one embodiment of the present disclosure, we propose an RPT structure that defines an r value by considering the statistical characteristics of a residual block, and derives a residual block of the size of an existing transform block from a residual block of a reduced size determined according to the defined r value. If an additional transformation (i.e., a secondary transform) is applied to predict the statistical distribution of primary transform coefficients, quantized non-zero coefficients may be concentrated in a relatively low frequency region because a quantization process is applied to the primary transform coefficients. Accordingly, a reduced secondary transform for the statistical distribution of primary transform coefficients can define the statistical characteristics of the primary transform coefficients relatively simply by setting an r value for a given low frequency region. However, the RPT according to the present disclosure is fundamentally different from the reduced secondary transform in that it is a technique for defining an r value by considering the statistical characteristics of samples within a residual block that have very different characteristics from the distribution of the primary transform coefficients. Below, we describe various embodiments for determining the RPT kernel, which is a reduced-dimensional transformation matrix. In other words, a method for determining or defining the r value in the RPT is described below. In one embodiment of the present disclosure, the r value in the RPT can be determined by considering the worst case complexity allowed by the conversion system. In one embodiment, the worst case complexity can be calculated based on the number of multiplications per sample. Applying the RPT in both the forward and backward directions based on an M x N block requires MN * r multiplications. Since a 2D block is composed of a total of MN samples, the number of multiplications per sample can be calculated as (MN * r) / MN = r. Therefore, the r value can be configured to be less than or equal to the maximum number of multiplications per sample allowed. For example, if the maximum possible number of multiplications per sample is set to 16 for a 16x16 block, the r value can be determined to be less than or equal to 16. That is, the forward RPT kernel can be set to 16 x 256. In another embodiment, memory usage can be considered as a measure of worst-case complexity. As an example, the allowed memory size per kernel can be set. For example, if each kernel coefficient (each element constituting a transform kernel is referred to as a kernel coefficient in the present disclosure) requires p bytes and the memory usage per kernel is set to be q bytes or less, the value of r can be set to be q / (MN * p) or less. For example, if p is 1 byte for a forward RPT kernel for a 16x16 block and the memory usage per kernel is set to be 8KB or less (q = 8KB = 2 13 bytes), the r value can be set to 32 or less. Additionally, as another example, memory usage and / or the number of multiplications per sample can be considered as measures of worst-case complexity. For example, if the maximum possible number of multiplications per sample is set to 16 for a 16x16 block and memory usage is set to 8KB or less per kernel (kernel coefficients are expressed as 1 byte), the value of r can be set to 16 or less. In addition, in one embodiment, the r value constituting the RPT kernel may be determined by specific information. In other words, the r value constituting the RPT kernel may be determined based on a predefined encoding parameter. For example, the r value may be determined based on the size of a block. In other words, the RPT kernel may be variably determined based on the size of the block. Here, the block may be at least one of a coding block, a transform block, and a prediction block. In addition, for example, the r value may be determined based on prediction information. Here, the prediction information may include information regarding inter / intra prediction, intra prediction mode information, etc. In addition, for example, the r value may be determined based on signaled information (values ​​of syntax elements). For example, the r value may be variably determined based on a quantization parameter value. In addition, in terms of improving complexity, a predefined fixed value may be used as the r value, and the predefined fixed value may be determined based on signaled information. When the RPT kernel rx MN is multiplied by the sample signal, r transform coefficients can be obtained. The obtained r transform coefficients can be arranged according to a predefined scan order of the transform coefficients (e.g., forward / backward zig-zag scan order, forward / backward horizontal scan order, forward / backward vertical scan order, forward / backward diagonal scan order, a scan order specified based on an intra prediction mode, etc.). When the transform coefficients obtained from the forward RPT application are arranged according to such a scan order (e.g., a scan order per coefficient group (CG) unit can also be applied), if the value of r is smaller than MN, the inside of the M x N block cannot be completely filled with the r transform coefficients, and thus a blank space may be generated. In an embodiment of the present disclosure, the above-described blank space can be predicted in the following manner in consideration of the characteristics of the residual signal. - The values ​​of empty spaces can be filled using the values ​​of available surrounding pixels. - The values ​​in empty spaces can be filled based on the values ​​of available surrounding pixels and the intra prediction mode. For example, the values ​​in empty spaces can be predicted by performing intra prediction based on the values ​​of available surrounding pixels and the intra prediction mode. - You can fill in the empty space with a predefined fixed value (e.g. 0). - The empty space can be filled with values ​​from available surrounding pixels using a predefined intra prediction mode (e.g., planar mode). In the present disclosure, filling the empty space with 0 among the examples described above may be referred to as a zero-out process. When filling the empty space with 0, the following embodiment may be applied. If a non-zero transform coefficient is detected (or parsed) in the empty space portion during parsing of the transform coefficients on the decoding device side, it may be considered (or inferred) that the RPT is not applied. In other words, if a non-zero transform coefficient exists in the predefined area representing the empty space, it may be considered that the RPT is not applied. In this case, signaling (or parsing) for a flag indicating whether the RPT is applied and / or an index designating one of a plurality of RPT kernel candidates may not be performed. As an example, if a non-zero transform coefficient exists in the predefined area representing the empty space, a predefined variable value may be updated, and it may be inferred that the RPT is not applied based on the updated variable value. In one embodiment of the present disclosure, whether or not to apply RPT may be determined based on the size and / or shape of a block. Furthermore, the RPT kernel may be variably determined based on the size and / or shape of the block. Since the r value may vary based on the size and / or shape of the block (i.e., for each M x N block), the empty space may vary based on the size and / or shape of the block. Accordingly, the area for checking whether a non-zero transform coefficient is detected may be defined differently based on the size and / or shape of the block. In other words, the zero-out area may be variably determined. For example, when a 16x64 matrix is ​​applied as a forward RPT matrix for an 8x8 block, the r value may be 16. In this case, when the CG is a 4x4 sub-block, only the upper left 4x4 block may be filled with a non-zero RPT transform coefficient, and the remaining three empty 4x4 sub-blocks (i.e., the upper right, lower left, and lower right sub-blocks) may be filled with 0 values. In this case, if a non-zero transform coefficient is detected in the remaining three 4x4 sub-block areas during the decoding process, it may be considered that the RPT is not applied. In addition, as described above, a flag indicating whether to apply the RPT or an index specifying one of multiple RPT kernel candidates may not be signaled. Also, as an example, when a 32x128 matrix is ​​applied as a forward RPT matrix for a 16x8 block (i.e., the r value is 32), and the CG is a 4x4 sub-block, only two CGs in the scan order can be filled with non-zero RPT transform coefficients. For example, the RPT transform coefficients can be filled in the upper-left 4x4 sub-block and the 4x4 sub-block adjacent to the lower side of the upper-left sub-block. The area to be filled with zero as an empty space can be determined as the remaining area excluding the two 4x4 sub-blocks. The RPT kernel can be determined variably depending on the size and / or shape of the block, and as seen, the empty space can be determined differently for an 8x8 block and a 16x8 block. As an example, if the value of r is a multiple of the CG size and the transform coefficients are scanned in CG units, the flag and / or index associated with the RPT may not be signaled if it is detected that non-zero transform coefficients exist in CGs belonging to empty spaces. That is, the transform coefficients within each CG may be scanned in a specified order, and the transform coefficients within the CG may be scanned in the same manner for the next CG according to the scan order for the CG unit. In conventional image compression techniques, since a flag for whether a non-zero transform coefficient exists within each CG is signaled first, whether to apply the RPT can be determined based on that information alone, which can reduce signaling overhead and related implementation complexity. As described above, if a non-zero transform coefficient is detected in an empty space area that is filled with 0 when RPT is applied, RPT may not be applied. In this case, signaling for information related to RPT may be omitted. However, since it is not possible to determine whether RPT is applied if a non-zero transform coefficient is not detected in the empty space area, the flag indicating whether RPT is applied can be parsed after parsing (or signaling) the related transform coefficients to make a final determination as to whether RPT is applied. In one embodiment, a forward secondary transform may be additionally applied to the transform coefficients generated through the application of RPT. Alternatively, a forward secondary transform may be additionally applied to an area where the generated transform coefficients are located within an M x N block. In the present disclosure, the area or a part of the area may be referred to as an ROI from the perspective of the forward secondary transform. In the reverse direction, the backward secondary transform may be applied first and then the backward RPT may be applied. Specifically, an area or a part of the area where r transform coefficients generated by the application of the forward RPT are located may be set as an ROI and the forward secondary transform may be applied. In this case, when a 16x64 forward RPT transform matrix is ​​applied to an 8x8 area, the 16 generated transform coefficients may be located in the upper left 4x4 sub-block, and the sub-block area may be set as an ROI and the forward secondary transform may be applied to the ROI. In addition, the coefficient values ​​of the RPT kernel can be adjusted considering operations such as integer arithmetic or fixed-point arithmetic. That is, the RPT kernel can be configured to perform a transformation through integer arithmetic (or fixed-point arithmetic) in a practical codec system by appropriately scaling the kernel coefficients belonging to the kernel, rather than a theoretical orthogonal transformation or non-orthogonal transformation (wherein, orthogonal transformation and non-orthogonal transformation refer to transformations in which the norm of each transformation basis vector is 1). When applying a separable transformation in a conventional image compression technology, the scaling factor multiplied can be equally reflected when applying the RPT. In this case, a separable transformation or a non-separable transformation (including RPT) can be performed while maintaining other processes (e.g., quantization and dequantization processes) other than the transformation. Integer coefficients of the RPT kernel can be obtained by multiplying the transformation basis vector by the scaling value described above. In an embodiment, multiplying the scaling value may include applying operations such as rounding, flooring, and ceilinging to each kernel coefficient. That is, the integerized RPT kernel obtained through the above-described method is defined and can be used in the transformation / inverse transformation process. As described above, if kernel coefficients whose integer values ​​are scaled through operations such as rounding, flooring, and ceilinging are obtained, the maximum and minimum values ​​can be obtained for all kernel coefficients, and thus the number of bits that can express all kernel coefficients can be obtained from the maximum and minimum values. For example, if the maximum value is less than or equal to 127 and the minimum value is greater than or equal to -128, all integer kernel coefficients can be expressed with 8 bits (especially, through 2's complement representation, etc.). In general, the maximum value is (2 (N-1) - 1) and the minimum value is -2 (N-1) If this is the case, all integer kernel coefficients can be expressed with N bits. If the maximum value is (2 (N-1) - 1) Greater than or equal to the minimum value -2 (N-1) For smaller cases, it may not be possible to represent all integer kernel coefficients with N bits. In this case, 1) all kernel coefficients can be additionally multiplied by a scaling value to fit within the N-bit range, or 2) the number of bits required to represent the kernel coefficients can be increased (i.e., more than N+1 bits). If all kernel coefficients are multiplied by 2 to represent them with N bits, -p If we need to multiply by that amount (p >= 1), then 2 is added to the existing encoding / decoding process. p Compensation can be made by multiplying the amount. As an example, 2 pMultiplying by can be implemented by performing an additional shift operation to the left by p bits, or by reducing the amount of right shift applied during quantization or dequantization by p. All kernel coefficients can be expressed as 8-bit, 9-bit, 10-bit, etc. using the method described above. Of course, the scaling value of the kernel coefficients can be set differently for each block size or kernel, and the number of bits for expressing the kernel coefficients can be set differently. The aforementioned NSPT can be applied based on at least one of the size, tree type, or component type of the current block. For example, whether to apply NSPT can be determined based on at least one of the size, tree type, or component type of the current block. An NSPT index can be signaled based on at least one of the size, tree type, or component type of the current block. An NSPT set or NSPT kernel can be derived based on at least one of the size, tree type, or component type of the current block. Allowed transform block sizes pre-defined in a decoding device can be broadly divided into two groups. One of the two groups (hereinafter referred to as the first group) may refer to a set of block sizes to which NSPT can be applied. The first group may be composed of any one of the allowed transform block sizes, or may be composed of two or more block sizes among the allowed block sizes. A block size to which NSPT can be applied may be defined as a block size in which at least one of the width and the height is less than or equal to a predetermined threshold. Alternatively, a block size to which NSPT can be applied may be defined as a block size in which the product of the width and the height is less than or equal to a predetermined threshold. Alternatively, a block size to which NSPT can be applied may be defined as a block size in which the maximum value of the width and the height is less than or equal to a predetermined threshold. The threshold may be an integer of 4, 8, 16, 32, 64, 128, or higher. The other of the two groups above (hereinafter referred to as the "second group") may refer to a set of block sizes to which NSPT is not applied. The aforementioned separable primary transformation may be applied to the block sizes belonging to the second group. Furthermore, a non-separable secondary transformation may be applied to all or some of the block sizes belonging to the second group. For example, if the size of the current block belongs to the first group, the reverse NSPT may be applied to the (dequantized) transform coefficients of the current block. If the size of the current block belongs to the second group, the reverse separable primary transform may be applied to the (dequantized) transform coefficients of the current block. Alternatively, if the size of the current block belongs to the second group, the reverse non-separable secondary transform (e.g., low frequency non-separable transform, LFNST) may be first applied to the (dequantized) transform coefficients of the current block, and then the reverse separable primary transform (e.g., DCT-2) may be applied to the transform coefficients obtained therefrom. For example, the first group, which is a set of block sizes to which NSPT can be applied, can be defined as a set of 4x4, 4x8, 8x4, 8x8. Alternatively, the first group can be defined as a set of 4x8, 8x4, 8x8. Alternatively, the first group can be defined as a set of 4x8, 8x4. Alternatively, the first group can be defined as a set of 4x4, 4x8, 4x16, 8x4, 8x8, 16x4. Alternatively, the first group can be defined as a set of 4x8, 4x16, 8x4, 8x8, 16x4. Alternatively, the first group can be defined as a set of 4x8, 4x16, 8x4, 16x4. Alternatively, the first group can be defined as a set of 4x4, 4x8, 8x4, 8x8, 8x16, 16x8, 16x16. Alternatively, the first group can be defined as a set of 4x4, 4x8, 8x4, 8x8, 8x16, 16x8. Alternatively, the first group can be defined as a set of 4x8, 8x4, 8x8, 8x16, 16x8. Alternatively, the first group can be defined as a set of 4x4, 4x8, 8x4, 8x8, 8x16, 16x8. Alternatively, the first group can be defined as a set of 4x4, 4x8, 8x4, 8x8, 8x16, 16x8, 16x16, 16x32, 32x16, 32x32. Alternatively, the first group can be defined as the set of 4x4, 4x8, 8x4, 8x8, 8x16, 16x8, 16x16, 16x32, 32x16. Alternatively, the first group can be defined as the set of 4x8, 8x4, 8x8, 8x16, 16x8, 16x16, 16x32, 32x16. Alternatively, the first group can be defined as the set of 4x8, 8x4, 8x16, 16x8, 16x16, 16x32, 32x16. Alternatively, the first group can be defined as the set of 4x8, 8x4, 8x16, 16x8, 16x32, 32x16. Alternatively, the first group can be defined as a set of 4x4, 4x8, 4x16, 8x4, 16x4.Alternatively, the first group can be defined as a set of 4x4, 4x8, 4x16, 8x4, 8x8, 8x16, 16x4, 16x8. Alternatively, the first group can be defined as a set of 4x8, 4x16, 8x4, 8x8, 8x16, 16x4, 16x8. Alternatively, the first group can be defined as a set of 4x4, 4x8, 4x16, 8x4, 8x16, 16x4, 16x8. Alternatively, the first group can be defined as a set of 4x8, 4x16, 8x4, 8x16, 16x4, 16x8. Alternatively, the first group can be defined as the set of 4x4, 4x8, 4x16, 8x4, 8x8, 8x16, 16x4, 16x8, 16x16. Alternatively, the first group can be defined as the set of 4x8, 4x16, 8x4, 8x8, 8x16, 16x4, 16x8, 16x16. Alternatively, the first group can be defined as the set of 4x4, 4x8, 4x16, 8x4, 8x16, 16x4, 16x8, 16x16. Alternatively, the first group can be defined as the set of 4x8, 4x16, 8x4, 8x16, 16x4, 16x8, 16x16. Alternatively, the first group can be defined as the set of 4x4, 4x8, 4x16, 4x32, 8x4, 8x16, 16x4, 16x8, 32x4, 32x8. Alternatively, the first group can be defined as the set of 4x8, 4x16, 4x32, 8x4, 8x16, 8x32, 16x4, 16x8, 32x4, 32x8. The first group can be defined as the set of 4x4, 4x8, 4x16, 4x32, 8x4, 8x8, 8x16, 8x32, 16x4, 16x8, 16x16, 32x4, 32x8. Alternatively, the first group can be defined as a set of 4x4, 4x8, 4x16, 4x32, 8x4, 8x8, 8x16, 8x32, 16x4, 16x8, 16x32, 32x4, 32x8, 32x16.Alternatively, the first group can be defined as the set of 4x4, 4x8, 4x16, 4x32, 8x4, 8x8, 8x16, 8x32, 16x4, 16x8, 16x16, 16x32, 32x4, 32x8, 32x16. Alternatively, the first group can be defined as the set of 4x8, 4x16, 4x32, 8x4, 8x16, 8x32, 16x4, 16x8, 16x32, 32x4, 32x8, 32x16. Alternatively, the first group can be defined as a set of 4x4, 4x8, 4x16, 4x32, 8x4, 8x8, 8x16, 8x32, 16x4, 16x8, 16x16, 16x32, 32x4, 32x8, 32x16, 32x32. For block sizes belonging to the first group, an NSPT matrix (or NSPT kernel) having a predetermined dimension can be applied. Here, the NSPT matrix can be expressed as a matrix having a dimension of PxQ as an inverse transformation matrix, and the PxQ matrix represents a matrix in which the number of rows and the number of columns are P and Q, respectively. For example, for a 4x4 block, a 16x16 NSPT matrix may be applied. For at least one of a 4x8 block or an 8x4 block, a 32x20 NSPT matrix may be applied. For at least one of a 4x16 block or a 16x4 block, a 64x24 NSPT matrix may be applied. For an 8x8 block, a 64x32 NSPT matrix may be applied. For at least one of an 8x16 block or a 16x8 block, a 128x40 NSPT matrix may be applied. For a 16x16 block, a 256x44 NSPT matrix may be applied. For a 4x32 block or a 32x4 block, a 128x36, 128x38, or 128x40 NSPT matrix may be applied. For 8x32 blocks or 32x8 blocks, a 256x48 NSPT matrix can be applied. For 16x32 blocks or 32x16 blocks, a 512x52 or 512x54 NSPT matrix can be applied. Since the above PxQ matrix is ​​a reverse NSPT matrix, a Px1 output vector can be obtained by applying the PxQ matrix to a Qx1 input vector (i.e., (PxQ matrix) x (Qx1 input vector)). Here, the Qx1 input vector may correspond to (inverse quantized) transform coefficients in the current block to which NSPT is applied. At this time, the value of Q may mean the number of transform coefficients to which NSPT is applied, and may be less than or equal to the product of the width and the height of the current block. The value of Q may be variably determined based on the size of the current block among the block sizes belonging to the first group described above. Alternatively, the value of Q may be set identically for the block sizes belonging to the first group. The Px1 output vector may correspond to a residual signal (or decoded residual samples). The value of P may be equal to the product of the width and the height of the current block. Conversely, the forward NSPT matrix can be expressed as a QxP matrix, which is the transpose of the PxQ matrix. A QxP output vector can be obtained by applying the QxP matrix to a Px1 input vector (i.e., (QxP matrix)x(Px1 input vector)). Here, the Px1 input vector can correspond to residual samples in the current block to which NSPT is applied. The value of P can be equal to the product of the width and the height of the current block. The Qx1 output vector can correspond to transform coefficients in the current block derived through NSPT. At this time, the value of Q can mean the number of transform coefficients output through NSPT, and can be less than or equal to the product of the width and the height of the current block. Similarly, the value of Q can be variably determined based on the size of the current block among the block sizes belonging to the first group described above. Alternatively, the value of Q can be set identically for the block sizes belonging to the first group. As in the above example, NSPT can be applied to non-square blocks, such as MxN blocks and NxM blocks. For example, NSPT can be applied to 4x8 blocks and 8x4 blocks. Alternatively, NSPT can be applied to 4x16 blocks and 16x4 blocks, or 8x16 blocks and 16x8 blocks, or 16x32 blocks and 32x16 blocks. By applying NSPT to specific block sizes belonging to the first group, more precise transformation can be performed, and coding performance can be improved. When forward LFNST is applied, the primary transformed transform coefficients of the remaining regions except for the region to which LFNST is applied (i.e., Region-Of-Interest, ROI) can be zeroed out. In addition, LFNST can be composed of a small number of transform basis vectors. In this case, if a separable primary transform such as DCT-2 and a non-separable secondary transform such as LFNST are applied instead of NSPT to the corresponding block sizes, performance degradation may occur. In this case, if NSPT is applied instead of LFNST, the zeroing process is omitted, and coding performance can be improved compared to when LFNST is applied. In addition, performance improvement can also be expected by applying NSPT. NSPT or LFNST can be applied using symmetry, which will be described later. Here, LFNST performs a transpose operation on the input block only for the ROI region, utilizing symmetry. In contrast, NSPT performs a transpose operation on the entire block, utilizing symmetry. Therefore, NSPT can train and apply the NSPT kernel using a more sophisticated symmetry method, which is expected to improve performance. Also, when applying LFNST instead of NSPT to an 8x8 block, a 32x64 transformation matrix can be applied instead of a 16x64 transformation matrix from the forward transformation perspective. Here, the 16x64 transformation matrix can be constructed by sampling the upper 16 rows of the 32x64 transformation matrix. When applying LFNST based on the 16x64 transformation matrix to an 8x8 block, 16 multiplications are required per sample to apply LFNST, but when using a 32x64 transformation matrix, 32 multiplications are required per sample to apply LFNST. However, when using a 32x64 transformation matrix in this way, improvement in coding performance can be expected. If the tree type of the current block is a single tree, NSPT can be applied to the luma component of the current block, but NSPT can not be applied to the chroma component of the current block. If the tree type of the current block is a dual tree, NSPT can be applied to the luma and chroma components of the current block. Alternatively, NSPT may be applied to the luma component of the current block and not applied to the chroma component of the current block, regardless of whether the tree type of the current block is a single tree. Alternatively, NSPT may be applied to the luma component and chroma component of the current block, regardless of whether the tree type of the current block is a single tree. For example, if the tree type of the current block is single tree, NSPT is allowed for luma and chroma components, and the size of the current block belongs to the first group, one NSPT index may be signaled, and the luma and chroma components of the current block may share the NSPT index. Here, the NSPT index may be an index for selecting any one of the transformation kernel candidates for NSPT. If the sizes of the luma block and the chroma block of the current block belong to the first group, the transformation kernel candidate selected by the same NSPT index may be applied to the luma and chroma components. If the tree type of the current block is single tree and NSPT is applied only to the luma component, LFNST may not be applied to the chroma component of the current block, and a separate transformation may be applied. Alternatively, if the tree type of the current block is single tree and NSPT is applied only to the luma component, LFNST may be applied to the chroma component of the current block. In the case of a single tree, the luma and chroma components may have highly correlated characteristics. In this case, applying NSPT only to the luma component or applying a transform kernel candidate selected by a single NSPT index to both luma and chroma components can reduce unnecessary signaling and improve compression efficiency. On the other hand, in the case of a non-single tree, the luma and chroma components each have independent partitioning and encoding structures. In this case, signaling the NSPT index for each component can reflect the characteristics of each component and improve compression efficiency. The NSPT kernel for the above NSPT can be derived based on at least one of symmetry between intra prediction modes or symmetry between block shapes. For example, the NSPT kernel can be derived as an NSPT kernel corresponding to at least one of a mode symmetrical to the intra prediction mode of the current block or a block shape symmetrical to the block shape of the current block. Alternatively, the NSPT kernel can be derived based on an NSPT set including one or more NSPT kernel candidates, wherein the NSPT set can be derived as an NSPT set corresponding to at least one of a mode symmetrical to the intra prediction mode of the current block or a block shape symmetrical to the block shape of the current block. Any one of the one or more NSPT kernel candidates belonging to the NSPT set can be set as the NSPT kernel of the current block. For this purpose, an NSPT index specifying any one of the one or more NSPT kernel candidates belonging to the NSPT set can be used. The NSPT index can be signaled through the bitstream or can be derived based on the aforementioned symmetry. There may be symmetry between at least two intra prediction modes among the intra prediction modes pre-defined in the decoding device. For convenience of explanation, the symmetry will be described centered on the upper left diagonal mode (i.e., mode 34). Referring to Fig. 5, there is symmetry between the directional modes. All modes have a prediction direction except for the Planar mode (number 0) and the DC mode (number 1). Modes 2 to 66 are named normal directional modes (which can be expressed as [2, 66]), and modes -14 to -1 (which can be expressed as [-14, -1]) and modes 67 to 80 (which can be expressed as [67, 80]) are named wide directional modes. The wide directional modes may include at least one of a mode having a value smaller than -14 or a mode having a value larger than 80. Referring to Fig. 5, all modes except modes 0 and 1 are symmetrical around mode 34. Specifically, for mode [2, 66], mode x and mode (68 - x) are symmetrical, and mode x and mode (66 - x) are symmetrical between mode [-14, -1] and mode [67, 80]. The same symmetry relationship can be established between mode [N, -1] and mode [67, 66 - N]. Here, N can be an integer less than or equal to -14. Meanwhile, with respect to the symmetry between block shapes, an MxN block and an NxM block can be defined as blocks that are symmetric to each other. Here, M and N can be the same or different. Alternatively, if the ratio of the width and height of an M1xN1 block (M1 / N1) and the ratio of the height and width of an M2xN2 block (N2 / M2) are the same, an M1xN1 block and an M2xN2 block can be defined as blocks that are symmetric to each other. Alternatively, if the ratio of the width and height of an M1xN1 block (M1 / N1) and the ratio of the width and height of an M2xN2 block (M2 / N2) are the same, an M1xN1 block and an M2xN2 block can be defined as blocks that are symmetric to each other. In a square block, modes that are symmetric to each other can share at least one of the NSPT set, NSPT index, or NSPT kernel. That is, at least one of the NSPT set, NSPT index, or NSPT kernel for any one of the symmetric modes can be applied equally to any other one of the symmetric modes. For example, modes that are symmetric to each other can share a single NSPT kernel. However, for one of the symmetric modes, the corresponding NSPT kernel can be applied to the input data, and for the other mode, a transpose operation can be applied to the input data and then the corresponding NSPT kernel can be applied. Specifically, if mode x belongs to the [2, 33] mode, for mode x, a one-dimensional vector (1D vector) can be constructed in row-first order for the input data, which is an MxM block, and the NSPT kernel can be applied to the 1D vector. Here, the construction of the 1D vector in column-first order can be done by reading the input data in units of columns from the input data, which is an MxM block, obtaining M columns, and arranging them in order to construct the 1D vector. On the other hand, for the (68 - x) mode that is symmetric to mode x, a 1D vector can be constructed in column-first order and the same NSPT kernel can be applied to the 1D vector. Here, the configuration of a 1D vector according to the row-major order may be to read the input data row by row from the MxM block of input data, obtain M rows, and sequentially arrange them to configure a 1D vector. If the x-th mode belongs to the [N, -1] mode (N ≤ -14), a 1D vector may be configured according to the column-first order for the (66 - x) mode, which is symmetric to the x-th mode, and the same NSPT kernel as that of the x-th mode may be applied to the 1D vector. The column-major order or the row-major order may be applied to the 0th and 1st modes, and the column-major order or the row-major order may also be applied to the 34th mode.Additionally, row-major ordering may be applied to intra prediction modes belonging to the [2, 33] mode, and column-major ordering may be applied to modes symmetrical to the intra prediction mode. Row-major ordering may be applied to intra prediction modes belonging to the [N, -1] mode, and column-major ordering may be applied to modes symmetrical to it. For non-square blocks, in addition to the symmetry between intra prediction modes, the symmetry between block shapes can be further considered. A non-square block with width and height M and N, respectively, can be viewed as having a symmetry relationship with a non-square block with width and height N and M, respectively. For example, in the [2, 66] mode, there may be symmetry between the x-th mode of an MxN block and the (68 - x)-th mode of an NxM block. Similarly, if the x-th mode of an MxN block belongs to the [N, -1] mode (N ≤ -14), there may be symmetry between the x-th mode of the MxN block and the (66 - x)-th mode of an NxM block. The method of constructing a 1D vector from an input data block is as discussed above. That is, if a column-major order is applied to mode x, a row-major order can be applied to a mode that is symmetric to it. Alternatively, if a row-major order is applied to mode x, a column-major order can be applied to a mode that is symmetric to it. Specifically, if a column-major order is applied to mode x, input data can be read column-wise from an MxN block of input data to obtain M columns, which can be arranged sequentially to construct a 1D vector. Here, each column can have a length of N. For a mode that is symmetric to mode x, input data can be read row-wise from an MxN block of input data to obtain N rows, which can be arranged sequentially to construct a 1D vector. Here, each row can have a length of M. Alternatively, if a row-major order is applied to mode x, input data can be read row-wise from an MxN block of input data to obtain N rows, which can be arranged sequentially to construct a 1D vector. Here, each row can have a length of M. For a mode symmetric to mode x, input data can be read column by column from an MxN block of input data to obtain M columns, which can be arranged in order to form a 1D vector. Here, each column can have a length of N. If the current block is an MxN block with mode x and the aforementioned symmetry is utilized for the current block, the NSPT set and / or NSPT kernel of the current block may be determined based on at least one of an intra prediction mode symmetric to mode x or a block size of NxM symmetric to the block size of MxN. Here, the NSPT kernel may be set as an NSPT kernel for an NxM block, not an NSPT kernel for an MxN block. That is, if symmetry is utilized for the current block, the NSPT set and / or NSPT kernel for a block that has symmetry with the current block may be utilized in the same manner. As described above, a 1D vector may be constructed from an input data block according to a predetermined priority order, which may correspond to an input of the NSPT kernel. In addition, the symmetry may be restricted to be utilized only when the value of the intra prediction mode of the current block is greater than 34. That is, when the value of the intra prediction mode of the current block is greater than 34, a transpose operation may be applied when constructing a 1D vector from an input data block, and an NSPT set or NSPT kernel corresponding to a mode and / or block shape having symmetry with the current block may be utilized. Specifically, when the intra prediction mode of the current block belongs to the [N, -1] mode and the [2, 34] mode, symmetry may not be utilized for the current block. On the other hand, when the intra prediction mode of the current block belongs to the [35, 66] mode and the [67, 66 - N] mode, symmetry may be utilized for the current block. Here, N may be an integer less than or equal to -14. The derivation of the NSPT set or NSPT kernel based on the above symmetry can be adaptively performed based on the size of the current block. For example, for 4x4 blocks and 8x8 blocks, the NSPT set or NSPT kernel may be derived based on symmetry, and for 4x8 blocks and 8x4 blocks, the NSPT set or NSPT kernel may not be derived based on symmetry. Depending on whether the above symmetry is utilized, the number of available NSPT sets may vary. For example, if symmetry is utilized, the number of available NSPT sets may be 35, and if symmetry is not utilized, the number of available NSPT sets may be 67. Table 9 below shows an example in which the NSPT set is determined using symmetry, and shows the mapping relationship between the intra prediction mode and the NSPT set when the number of available NSPT sets is 35. Intra prediction mode NSPT set index X < 020 ≤ X ≤ 34 X35 ≤ X ≤ 6668 - XX > 662 Referring to Table 9, if the value (X) of the intra prediction mode of the current block is less than 0, the NSPT set of the current block may be determined as an NSPT set having an NSPT set index of 2 among 35 NSPT sets. If the value (X) of the intra prediction mode of the current block is greater than or equal to 0 and less than or equal to 34, the NSPT set of the current block may be determined as an NSPT set having an NSPT set index of X among 35 NSPT sets. If the value (X) of the intra prediction mode of the current block is greater than or equal to 35 and less than or equal to 66, the NSPT set of the current block may be determined as an NSPT set having an NSPT set index of (68-X) among 35 NSPT sets. If the value (X) of the intra prediction mode of the current block is greater than or equal to 35 and less than or equal to 66, the NSPT set of the current block may be identical to the NSPT set corresponding to the value (68-X) of the mode symmetrical to the intra prediction mode of the current block. Similarly, if the value (X) of the intra prediction mode of the current block is greater than 66, the NSPT set of the current block may be determined as the NSPT set having the NSPT set index of 2 among the 35 NSPT sets. If the value (X) of the intra prediction mode of the current block is greater than 66, the NSPT set of the current block may be identical to the NSPT set corresponding to the mode symmetrical to the intra prediction mode of the current block. Table 10 below shows an example in which the NSPT set is determined without using symmetry, and shows the mapping relationship between the intra prediction mode and the NSPT set when the number of available NSPT sets is 67. Intra prediction mode NSPT set index X < 020 ≤ X ≤ 66XX > 6666 Referring to Table 10, if the value (X) of the intra prediction mode of the current block is less than 0, the NSPT set of the current block may be determined as an NSPT set having an NSPT set index of 2 among 67 NSPT sets. If the value (X) of the intra prediction mode of the current block is greater than or equal to 0 and less than or equal to 66, the NSPT set of the current block may be determined as an NSPT set having an NSPT set index of X among 67 NSPT sets. If the value (X) of the intra prediction mode of the current block is greater than 66, the NSPT set of the current block may be determined as an NSPT set having an NSPT set index of 66 among 67 NSPT sets. By leveraging this symmetry, the memory required to store transformation kernels can be reduced while maintaining the performance required for transform application. For example, by utilizing symmetry to use 35 NSPT sets instead of 67, the memory required to store NSPT kernels can be significantly reduced. The number of available NSPT sets and / or the number of NSPT kernel candidates in an NSPT set may vary depending on the block size. For example, the number of available NSPT sets for a 4x4 block may be 35, the number of available NSPT sets for a 4x8 block and an 8x4 block may be 19, and the number of available NSPT sets for an 8x8 block may be 10. The NSPT set for a 4x4 block may consist of three NSPT kernel candidates, the NSPT sets for a 4x8 block and an 8x4 block may consist of three or two NSPT kernel candidates, and the NSPT set for an 8x8 block may consist of one NSPT kernel candidate. As the block size increases, the size of the transform kernel may increase. Therefore, by reducing the number of available NSPT sets and / or the number of NSPT kernel candidates within an NSPT set, the memory required to store the transform kernel can be saved. Furthermore, as the block size increases, the residual signal characteristics within the block tend to become more generalized. Therefore, reducing the number of available NSPT sets and / or the number of NSPT kernel candidates within an NSPT set can help maintain compression efficiency while reducing implementation complexity by reflecting these statistical characteristics. Example 4 When geometric partitioning is applied to a current block, the current block may be partitioned into two or more partitions. Here, the current block may be a coding block, a coding unit, a transform block, or a transform unit. Different prediction methods may be applied to each partition. For example, an inter mode may be applied to the first partition of the current block, and an intra mode may be applied to the second partition of the current block. Alternatively, an inter mode may be applied to the first and second partitions of the current block, but a merge mode may be applied to the first partition, and a different inter prediction mode (e.g., AMVP mode) other than the merge mode may be applied to the second partition. Alternatively, the same inter prediction mode (e.g., merge mode) may be applied to the first and second partitions of the current block, but different motion information may be applied to each partition. Alternatively, an intra mode may be applied to the first and second partitions of the current block, but different intra prediction modes may be applied to each partition. As described above, different prediction methods are applied to each partition belonging to a block, but when a single inverse transformation is applied to the entire block, a method for determining a transformation set and / or transformation kernel for the inverse transformation is proposed. A separate set of transformations for non-separable transformations that can be applied when geometric partitioning is applied to the current block may be introduced. Here, the set of transformations may consist of multiple transformation kernel candidates, and a non-separable transformation-based inverse transformation may be performed based on any one of the transformation kernel candidates. For this purpose, an index indicating any one of the transformation kernel candidates may be separately signaled. Additionally, a separate MTS set for a separate separable transformation that can be applied when geometric partitioning is applied to the current block can be introduced. The MTS set can be composed of multiple transformation kernel candidates, and a separable transformation-based inverse transformation can be performed based on any one of the transformation kernel candidates. For this purpose, an index indicating any one of the transformation kernel candidates belonging to the MTS set can be separately signaled. Each transformation kernel candidate can be defined as a pair of a horizontal transformation kernel and a vertical transformation kernel. For example, a pair of a horizontal transformation kernel and a vertical transformation kernel can be expressed as a 2-tuple, such as (DST-7, DCT-2). As described in the above-described embodiment, a transform set for non-separable transformation of a current block can be determined based on the intra prediction mode of the current block. Specifically, intra prediction modes predefined identically for an encoding device and a decoding device can be divided into multiple groups. Here, each group can be composed of one or more intra prediction modes. A transform set applicable to each group can be assigned. Based on the intra prediction mode of the current block, the group to which the intra prediction mode belongs can be determined, and then a transform set applicable to the group can be determined. When geometric partitioning is applied to the current block, a specific intra prediction mode is considered to be applied to the entire current block, and a transform set for non-separable transformation can be determined based on the specific intra prediction mode. Similarly, the MTS set for the separation transform can be determined based on the intra prediction mode of the current block. Furthermore, the MTS set can also be determined based on the intra prediction mode of the current block and its size. If geometric segmentation is applied to the current block, a specific intra prediction mode can be considered to be applied to the entire current block, and the MTS set can be determined based on that specific intra prediction mode. The intra prediction mode (hereinafter referred to as a virtual intra prediction mode) considered to be applied to the entire current block may be a pre-configured non-directional mode. For example, the pre-configured non-directional mode may be a Planar mode or a DC mode. If the virtual intra prediction mode is a Planar mode, a transform set for a non-separable transform corresponding to the Planar mode may be selected, or an MTS set corresponding to the Planar mode may be selected. Alternatively, the virtual intra prediction mode may be adaptively determined based on the segmentation direction according to the geometric segmentation, and a method for deriving the virtual intra prediction mode will be described in detail below. Method 1 An intra prediction mode that matches or is closest to the segmentation direction of the current block can be selected as the virtual intra prediction mode. A decoding device may predefine multiple segmentation types for geometric segmentation. The decoding device may define a table (hereinafter referred to as a first mapping table) for mapping relationships between direction indices and distance indices corresponding to each segmentation type. The direction index may refer to an index (angleIdx) indicating a segmentation direction for each segmentation type, and the distance index may refer to an index (distanceIdx) indicating a distance between a segmentation line of the corresponding segmentation type and the center of the current block. For example, the first mapping table may be defined as shown in Table 11 below. merge_gpm_partition_idx0123456789101112131415angleIdx0022223333444455distanceIdx1301230123012301merge_gpm_partitio n_idx16171819202122232425262728293031angleIdx5588111111111212121213131313distanceIdx2313012301230123merge_gpm_parti tion_idx32333435363738394041424344454647angleIdx14141414161618181819191920202021distanceIdx0123131231231231merge_g pm_partition_idx48495051525354555657585960616263angleIdx21212424272727282828292929303030distanceIdx2313123123123123 In Table 11, when angleIdx is 0, this means that the angle between the line perpendicular to the dividing line and the horizontal axis pointing to the right is 0 degrees, and the dividing line is a vertical line. As the value of angleIdx increases, the dividing line rotates counterclockwise. That is, the angle between the line perpendicular to the dividing line and the horizontal axis pointing to the right increases. When angleIdx is 8, 16, and 24, this means that the angles increasing counterclockwise from the horizontal axis pointing to the right are 90 degrees, 180 degrees, and 270 degrees, respectively. When angleIdx is 32, this means that the angle increasing counterclockwise from the horizontal axis pointing to the right is 360 degrees, which can be the same as when angleIdx is 0. In Table 11, distanceIdx represents the distance between the dividing line and the center of the current block, and can have any of the values ​​of 0, 1, 2, or 3. As the value of distanceIdx increases, the splitting line may be farther away from the center of the current block. For example, in Table 11, if merge_gpm_partition_idx is 10, 11, 12, and 13, distanceIdx may be 0, 1, 2, and 3, respectively, and the distances from the center may be 1, 3, 6, and 10, respectively. That is, as the value of distanceIdx increases, the distance from the center may increase evenly or unevenly. In Table 11, if distanceIdx is 0, this may mean that the distance from the center of the current block is 0 or very close. In Table 11, merge_gpm_partition_idx has values ​​from 0 to 63, a total of 64 split types are defined, and a splitting direction can be defined for each of the 64 split types. Based on the first mapping table, a direction index corresponding to the segmentation type of the current block can be determined. An intra prediction mode that matches or has the closest prediction direction to the segmentation direction according to the determined direction index can be determined. The determined intra prediction mode can be set as the virtual intra prediction mode of the current block. A decoding device may define a table (hereinafter referred to as a "second mapping table") for mapping relationships between segmentation directions (i.e., angleIdx) and intra prediction modes. Here, an intra prediction mode mapped to a specific segmentation direction may be an intra prediction mode that matches or has a prediction direction closest to the specific segmentation direction. For example, the second mapping table may be defined as shown in Table 12 below. g_geoAngle2IntraAng[GEO_NUM_ANGLES] = {50, 0, 44, 41, 34, 27, 0, 0, 18, 0, 0, 9, 66, 59, 56, 0, 50, 0, 44, 41, 34, 27, 0, 0, 18, 0, 0, 9, 66, 59, 56, 0}; Table 12 defines the intra prediction mode corresponding to each of the 32 segmentation directions. In Table 12, the numbers in parentheses represent the values ​​of the intra prediction modes. For example, if angleIdx is 0, 50 corresponding to g_geoAngle2IntraAng[0] is selected, and an intra prediction mode with a value of 50 can be derived. Alternatively, if angleIdx is 8, 18 corresponding to g_geoAngle2IntraAng[8] is selected, and an intra prediction mode with a value of 18 can be derived. Alternatively, if angleIdx is 31, g_geoAngle2IntraAng

[0031] 0 corresponding to is selected, and an intra prediction mode with a value of 0 can be derived. The prediction direction corresponding to the value of the intra prediction mode is as shown in Fig. 5. When geometric partitioning is applied to the current block, angleIdx corresponding to the partition type of the current block can be determined based on the first mapping table. An intra prediction mode corresponding to the angleIdx can be derived based on the second mapping table. The derived intra prediction mode can be set as a virtual intra prediction mode for the current block. For example, when merge_gpm_partition_idx is 11 in Table 11, the corresponding angleIdx is 4, so 34 corresponding to g_geoAngle2IntraAng[4] can be selected through the mapping table in Table 12. The intra prediction mode with the value 34 can be set as a virtual intra prediction mode for the current block. Method 2 An intra prediction mode that matches or is closest to the direction perpendicular to the segmentation direction of the current block may be selected as the virtual intra prediction mode. Alternatively, based on the second mapping table described above, an intra prediction mode corresponding to the direction perpendicular to the segmentation direction of the current block may be derived, and the derived intra prediction mode may be set as the virtual intra prediction mode for the current block. When a block is divided into two partitions, one of the partitions may be adjacent to a reference sample for intra prediction, but the accuracy of intra prediction from that reference sample tends to decrease. Therefore, if the partitioning boundary is set in a direction perpendicular to the prediction direction of the intra prediction mode, the partitions can be divided into different partitions and different intra prediction modes can be applied starting from a point that is somewhat distant from the prediction direction of the intra prediction mode. Therefore, selecting the intra prediction mode that matches or is closest to the direction perpendicular to the partitioning direction can be a method for applying the most appropriate intra prediction mode to each partition. Method 3 A virtual intra prediction mode for a current block can be determined based on the partition type of the current block. A partition index corresponding to the partition type of the current block can be derived. To this end, a separate mapping table (hereinafter referred to as a third mapping table) in which a partition index corresponding to each pre-defined partition type is defined can be used. The partition index can specify one of two partitions. The partition index can specify the location of a larger partition among the two partitions within the current block. A virtual intra prediction mode for the current block can be derived based on the intra prediction mode of the partition specified by the partition index. The third mapping table may be defined according to block shape. Here, the block shape may be divided into square blocks and non-square blocks. In this case, the third mapping table may be composed of a mapping table for square blocks and a mapping table for non-square blocks. If the current block is a square block such as 4x4 or 8x8, the mapping table for the square block may be used for the current block. On the other hand, if the current block is a non-square block such as 4x8 or 8x4, the mapping table for the non-square block may be used for the current block. Alternatively, the block shape may be classified into a square block, a fat block (when w is greater than h), and / or a thin block (when w is less than h) according to the width (w) and height (h) of the block. In this case, the third mapping table may be composed of a mapping table for a square block, a mapping table for a fat block, and a mapping table for a thin block. When the current block is a square block such as 4x4 or 8x8, the mapping table for the square block may be used for the current block. When the current block is a fat block such as 8x4 or 16x4, the mapping table for the fat block may be used for the current block. When the current block is a thin block such as 4x8 or 4x16, the mapping table for the thin block may be used for the current block. For example, the mapping tables for the forward block, fat block, and thin block can be defined as shown in Table 13 below. uint8_t g_geoMajorPartSquare[ GEO_NUM_PARTITION_MODE ] ={0, / 0: (angleIdx, distanceIdx) = ( 0, 1)0, / 1: (angleIdx, distanceIdx) = ( 0, 3)1, / 2: (angleIdx, distanceIdx) = ( 2, 0)1, / 3: (angleIdx, distanceIdx) = ( 2, 1)1, / 4: (angleIdx, distanceIdx) = ( 2, 2)1, / 5: (angleIdx, distanceIdx) = ( 2, 3)1, / 6: (angleIdx, distanceIdx) = ( 3, 0)1, / 7: (angleIdx, distanceIdx) = ( 3, 1)1, / 8: (angleIdx, distanceIdx) = ( 3, 2)1, / 9: (angleIdx, distanceIdx) = ( 3, 3)1, / 10: (angleIdx, distanceIdx) = ( 4, 0)1, / 11: (angleIdx, distanceIdx) = ( 4, 1)1, / 12: (angleIdx, distanceIdx) = ( 4, 2)1, / 13: (angleIdx, distanceIdx) = ( 4, 3)0, / 14: (angleIdx, distanceIdx) = ( 5, 0)1, / 15: (angleIdx, distanceIdx) = ( 5, 1)1, / 16: (angleIdx, distanceIdx) = ( 5, 2)1, / 17: (angleIdx, distanceIdx) = ( 5, 3)1, / 18: (angleIdx, distanceIdx) = ( 8, 1)1, / 19: (angleIdx, distanceIdx) = ( 8, 3)0, / 20: (angleIdx, distanceIdx) = (11, 0)1, / 21: (angleIdx, distanceIdx) = (11, 1)1, / 22: (angleIdx, distanceIdx) = (11,2)1, / 23: (angleIdx, distanceIdx) = (11, 3)1, / 24: (angleIdx, distanceIdx) = (12, 0)1, / 25: (angleIdx, distanceIdx) = (12, 1)1, / 26: (angleIdx, distanceIdx) = (12, 2)1, / 27: (angleIdx, distanceIdx) = (12, 3)1, / 28: (angleIdx, distanceIdx) = (13, 0)0, / 29: (angleIdx, distanceIdx) = (13, 1)0, / 30: (angleIdx, distanceIdx) = (13, 2)0, / 31: (angleIdx, distanceIdx) = (13, 3)1, / 32: (angleIdx, distanceIdx) = (14, 0)0, / 33: (angleIdx, distanceIdx) = (14, 1)0, / 34: (angleIdx, distanceIdx) = (14, 2)0, / 35: (angleIdx, distanceIdx) = (14, 3)1, / 36: (angleIdx, distanceIdx) = (16, 1)1, / 37: (angleIdx, distanceIdx) = (16, 3)0, / 38: (angleIdx, distanceIdx) = (18, 1)0, / 39: (angleIdx, distanceIdx) = (18, 2)0, / 40: (angleIdx, distanceIdx) = (18, 3)0, / 41: (angleIdx, distanceIdx) = (19, 1)0, / 42: (angleIdx, distanceIdx) = (19, 2)0, / 43: (angleIdx, distanceIdx) = (19, 3)0, / 44: (angleIdx, distanceIdx) = (20, 1)0, / 45: (angleIdx, distanceIdx) = (20, 2)0, / 46: (angleIdx, distanceIdx) = (20, 3)0, / 47: (angleIdx, distanceIdx) = (21, 1)0, / 48: (angleIdx, distanceIdx) = (21, 2)0, / 49: (angleIdx, distanceIdx) = (21, 3)0, / 50: (angleIdx, distanceIdx) = (24, 1)0, / 51: (angleIdx, distanceIdx) = (24, 3)0, / 52: (angleIdx, distanceIdx) = (27, 1)0, / 53: (angleIdx, distanceIdx) = (27, 2)0, / 54: (angleIdx, distanceIdx) = (27, 3)1, / 55: (angleIdx, distanceIdx) = (28, 1)1, / 56: (angleIdx, distanceIdx) = (28, 2)1, / 57: (angleIdx, distanceIdx) = (28, 3)1, / 58: (angleIdx, distanceIdx) = (29, 1)1, / 59: (angleIdx, distanceIdx) = (29, 2)1, / 60: (angleIdx, distanceIdx) = (29, 3)1, / 61: (angleIdx, distanceIdx) = (30, 1)1, / 62: (angleIdx, distanceIdx) = (30, 2)1, / 63: (angleIdx, distanceIdx) = (30, 3)};uint8_t g_geoMajorPartThin[ GEO_NUM_PARTITION_MODE ] ={0, / 0: (angleIdx, distanceIdx) = ( 0, 1)0, / 1: (angleIdx, distanceIdx) = ( 0, 3)1, / 2: (angleIdx, distanceIdx) = ( 2, 0)1, / 3: (angleIdx, distanceIdx) = ( 2, 1)1, / 4: (angleIdx, distanceIdx) = ( 2, 2)1, / 5: (angleIdx, distanceIdx) = ( 2, 3)1, / 6: (angleIdx, distanceIdx) = ( 3, 0)1, / 7: (angleIdx, distanceIdx) = ( 3, 1)1, / 8: (angleIdx, distanceIdx) = ( 3, 2)1, / 9: (angleIdx, distanceIdx) = ( 3, 3)1, / 10: (angleIdx, distanceIdx) = ( 4, 0)1, / 11: (angleIdx, distanceIdx) = ( 4, 1)1, / 12: (angleIdx, distanceIdx) = ( 4, 2)1, / 13: (angleIdx, distanceIdx) = ( 4, 3)0, / 14: (angleIdx, distanceIdx) = ( 5, 0)1, / 15: (angleIdx, distanceIdx) = ( 5, 1)1, / 16: (angleIdx, distanceIdx) = ( 5, 2)1, / 17: (angleIdx, distanceIdx) = ( 5, 3)1, / 18: (angleIdx, distanceIdx) = ( 8, 1)1, / 19: (angleIdx, distanceIdx) = ( 8, 3)0, / 20: (angleIdx, distanceIdx) = (11, 0)1, / 21: (angleIdx, distanceIdx) = (11, 1)1, / 22: (angleIdx, distanceIdx) = (11, 2)1, / 23: (angleIdx, distanceIdx) = (11, 3)1, / 24: (angleIdx, distanceIdx) = (12, 0)1, / 25: (angleIdx, distanceIdx) = (12, 1)1, / 26: (angleIdx, distanceIdx) = (12, 2)1, / 27: (angleIdx, distanceIdx) = (12, 3)1, / 28: (angleIdx, distanceIdx) = (13, 0)0, / 29: (angleIdx, distanceIdx) = (13, 1)0, / 30: (angleIdx,distanceIdx) = (13, 2)0, / 31: (angleIdx, distanceIdx) = (13, 3)1, / 32: (angleIdx, distanceIdx) = (14, 0)0, / 33: (angleIdx, distanceIdx) = (14, 1)0, / 34: (angleIdx, distanceIdx) = (14, 2)0, / 35: (angleIdx, distanceIdx) = (14, 3)1, / 36: (angleIdx, distanceIdx) = (16, 1)1, / 37: (angleIdx, distanceIdx) = (16, 3)0, / 38: (angleIdx, distanceIdx) = (18, 1)0, / 39: (angleIdx, distanceIdx) = (18, 2)0, / 40: (angleIdx, distanceIdx) = (18, 3)0, / 41: (angleIdx, distanceIdx) = (19, 1)0, / 42: (angleIdx, distanceIdx) = (19, 2)0, / 43: (angleIdx, distanceIdx) = (19, 3)0, / 44: (angleIdx, distanceIdx) = (20, 1)0, / 45: (angleIdx, distanceIdx) = (20, 2)0, / 46: (angleIdx, distanceIdx) = (20, 3)0, / 47: (angleIdx, distanceIdx) = (21, 1)0, / 48: (angleIdx, distanceIdx) = (21, 2)0, / 49: (angleIdx, distanceIdx) = (21, 3)0, / 50: (angleIdx, distanceIdx) = (24, 1)0, / 51: (angleIdx, distanceIdx) = (24, 3)0, / 52: (angleIdx, distanceIdx) = (27, 1)0, / 53: (angleIdx, distanceIdx) = (27, 2)0, / 54: (angleIdx,distanceIdx) = (27, 3)1, / 55: (angleIdx, distanceIdx) = (28, 1)1, / 56: (angleIdx, distanceIdx) = (28, 2)1, / 57: (angleIdx, distanceIdx) = (28, 3)1, / 58: (angleIdx, distanceIdx) = (29, 1)1, / 59: (angleIdx, distanceIdx) = (29, 2)1, / 60: (angleIdx, distanceIdx) = (29, 3)1, / 61: (angleIdx, distanceIdx) = (30, 1)1, / 62: (angleIdx, distanceIdx) = (30, 2)1, / 63: (angleIdx, distanceIdx) = (30, 3)};uint8_t g_geoMajorPartFat[ GEO_NUM_PARTITION_MODE ] ={0, / 0: (angleIdx, distanceIdx) = ( 0, 1)0, / 1: (angleIdx, distanceIdx) = ( 0, 3)1, / 2: (angleIdx, distanceIdx) = ( 2, 0)0, / 3: (angleIdx, distanceIdx) = ( 2, 1)0, / 4: (angleIdx, distanceIdx) = ( 2, 2)0, / 5: (angleIdx, distanceIdx) = ( 2, 3)1, / 6: (angleIdx, distanceIdx) = ( 3, 0)0, / 7: (angleIdx, distanceIdx) = ( 3, 1)0, / 8: (angleIdx, distanceIdx) = ( 3, 2)0, / 9: (angleIdx, distanceIdx) = ( 3, 3)1, / 10: (angleIdx, distanceIdx) = ( 4, 0)0, / 11: (angleIdx, distanceIdx) = ( 4, 1)0, / 12: (angleIdx, distanceIdx) = ( 4, 2)0, / 13: (angleIdx,distanceIdx) = ( 4, 3)0, / 14: (angleIdx, distanceIdx) = ( 5, 0)0, / 15: (angleIdx, distanceIdx) = ( 5, 1)0, / 16: (angleIdx, distanceIdx) = ( 5, 2)0, / 17: (angleIdx, distanceIdx) = ( 5, 3)1, / 18: (angleIdx, distanceIdx) = ( 8, 1)1, / 19: (angleIdx, distanceIdx) = ( 8, 3)0, / 20: (angleIdx, distanceIdx) = (11, 0)1, / 21: (angleIdx, distanceIdx) = (11, 1)1, / 22: (angleIdx, distanceIdx) = (11, 2)1, / 23: (angleIdx, distanceIdx) = (11, 3)1, / 24: (angleIdx, distanceIdx) = (12, 0)1, / 25: (angleIdx, distanceIdx) = (12, 1)1, / 26: (angleIdx, distanceIdx) = (12, 2)1, / 27: (angleIdx, distanceIdx) = (12, 3)1, / 28: (angleIdx, distanceIdx) = (13, 0)0, / 29: (angleIdx, distanceIdx) = (13, 1)0, / 30: (angleIdx, distanceIdx) = (13, 2)0, / 31: (angleIdx, distanceIdx) = (13, 3)1, / 32: (angleIdx, distanceIdx) = (14, 0)0, / 33: (angleIdx, distanceIdx) = (14, 1)0, / 34: (angleIdx, distanceIdx) = (14, 2)0, / 35: (angleIdx, distanceIdx) = (14, 3)1, / 36: (angleIdx, distanceIdx) = (16, 1)1, / 37: (angleIdx,distanceIdx) = (16, 3)1, / 38: (angleIdx, distanceIdx) = (18, 1)1, / 39: (angleIdx, distanceIdx) = (18, 2)1, / 40: (angleIdx, distanceIdx) = (18, 3)1, / 41: (angleIdx, distanceIdx) = (19, 1)1, / 42: (angleIdx, distanceIdx) = (19, 2)1, / 43: (angleIdx, distanceIdx) = (19, 3)1, / 44: (angleIdx, distanceIdx) = (20, 1)1, / 45: (angleIdx, distanceIdx) = (20, 2)1, / 46: (angleIdx, distanceIdx) = (20, 3)1, / 47: (angleIdx, distanceIdx) = (21, 1)1, / 48: (angleIdx, distanceIdx) = (21, 2)1, / 49: (angleIdx, distanceIdx) = (21, 3)0, / 50: (angleIdx, distanceIdx) = (24, 1)0, / 51: (angleIdx, distanceIdx) = (24, 3)0, / 52: (angleIdx, distanceIdx) = (27, 1)0, / 53: (angleIdx, distanceIdx) = (27, 2)0, / 54: (angleIdx, distanceIdx) = (27, 3)1, / 55: (angleIdx, distanceIdx) = (28, 1)1, / 56: (angleIdx, distanceIdx) = (28, 2)1, / 57: (angleIdx, distanceIdx) = (28, 3)1, / 58: (angleIdx, distanceIdx) = (29, 1)1, / 59: (angleIdx, distanceIdx) = (29, 2)1, / 60: (angleIdx, distanceIdx) = (29, 3)1, / 61: (angleIdx,distanceIdx) = (30, 1)1, / 62: (angleIdx, distanceIdx) = (30, 2)1, / 63: (angleIdx, distanceIdx) = (30, 3)};, Table 13 is a table showing the partition indices corresponding to the 64 partition types presented in Table 11 for square blocks, fat blocks, and thin blocks. The table in Table 13 follows the C / C++ syntax and is therefore described as a C / C++ array, and the index values ​​for accessing the array start from 0. In Table 13, GEO_NUM_PARTITION_MODE can be the total number of partition types for geometric partitioning (e.g., 64). g_geoMajorPartSquare, g_geoMajorPartThin, and g_geoMajorPartFat can be tables that indicate which partition is a larger partition for each of the 64 partition types for square blocks, thin blocks, and fat blocks, respectively. For example, if the current block is an 8x16 block, the current block corresponds to a thin block, so the g_geoMajorPartThin array can be used. At this time, if merge_gpm_partition_idx indicating the partition type of the current block is 41, 0 corresponding to g_geoMajorPartThin

[0041] can be output. This can indicate that the partition corresponding to index 0 is a larger partition. Through the above-described method, it can be determined which of the two partitions within the current block is the larger partition, and a virtual intra prediction mode for the current block can be derived based on the intra prediction mode of the larger partition. However, if the sizes of the partitions within the current block are the same, the intra prediction mode applied to any one of the partitions may be selected as the virtual intra prediction mode. Alternatively, if the sizes of the partitions within the current block are the same, a partition with a greater number of adjacent reference samples may be selected among the partitions, and the intra prediction mode of the selected partition may be selected as the virtual intra prediction mode. Here, the reference sample may belong to at least one of the left region, the upper region, or the upper left region of the current block. FIG. 6 illustrates an example of geometric partitioning according to the present disclosure. Currently, two partitions within a block can be distinguished by indices 0 and 1. Referring to FIG. 6, a P0 partition and a P1 partition can correspond to indices 0 and 1, respectively. The geometric partitioning illustrated in FIG. 6(b) can correspond to a case where the value of angleIdx is 13 or more and 27 or less, and the geometric partitioning illustrated in FIG. 6(a) can correspond to the remaining values ​​of angleIdx. Specifically, in the case of FIG. 6(a), a partition corresponding to index 0 can be located in the direction indicated by angleIdx (or, a direction corresponding to the angle indicated by angleIdx among directions perpendicular to the partitioning line), and the remaining partitions can correspond to index 1. In the case of Fig. 6(b), a partition corresponding to index 1 may be located in the direction indicated by angleIdx (or, in the direction corresponding to the angle indicated by angleIdx among the directions perpendicular to the dividing line), and the remaining partitions may correspond to index 0. Method 4 The virtual intra prediction mode for the current block can be derived as an intra prediction mode based on a decoder-side intra mode derivation (DIMD) method. Hereinafter, a method for deriving an intra prediction mode based on DIMD will be described in detail with reference to FIG. 7. Referring to FIG. 7, an initial sample position may be set, and accumulated strength values ​​for all intra prediction modes may be initialized (S700). A strength value for a current sample position may be calculated (S710). An intra prediction mode for which the calculated strength value is to be accumulated may be selected (S720). The strength value calculated in S710 may be added to the accumulated strength value for the selected intra prediction mode (S730). The processes of S710 to S730 described above may be performed for each of all or some sample positions belonging to the current block until a next sample position does not exist. If a next sample position does not exist, an intra prediction mode having the largest accumulated strength value may be selected (S740). The process of applying DIMD to the surrounding area of ​​the current block is described in detail as follows. For convenience of explanation, the surrounding area is assumed to be a block with width and height W and H. A filter of size PxQ can be applied inside a WxH block. The PxQ filter can be a 2D filter or a 1D filter. However, there may be cases where the filter goes beyond the block boundary for samples adjacent to the block boundary. The filter can be applied only to the internal area of ​​the WxH block excluding samples adjacent to the block boundary. Specifically, the filter can be applied only to sample positions belonging to the (W-P+1)x(H-Q+1) area, which is the internal area of ​​the WxH block. For example, when the filter size is 3x3, a 3x3 2D filter can be applied only to sample positions belonging to the (W-2)x(H-2) block excluding the edge of the length of 1 in the WxH block. In this case, the 2D filter can be applied to a 3x3 area having the sample of the sample position (hereinafter referred to as the reference sample) as the center sample. The above reference sample and at least one surrounding sample adjacent to the reference sample may be input to the 2D filter. Here, the surrounding sample may include a sample adjacent to at least one of the top, left, bottom, right, top left, top right, bottom left, or bottom right of the reference sample. A filter can be applied to each sample position belonging to the internal region to derive a strength value of a specific intra prediction mode. The derived strength value is added to a previously derived strength value for the specific intra prediction mode, and through this process, a cumulative strength value can be derived for the specific intra prediction mode. By applying a filter to all sample positions belonging to the internal region, cumulative strength values ​​can be derived for all intra prediction modes (or directional modes excluding the non-directional mode). An intra prediction mode with the largest cumulative strength value can be selected, and this can be set as an intra prediction mode derived based on DIMD (hereinafter, referred to as a DIMD mode). Two types of 3x3 filters, as shown in Fig. 8, can be used as filters to induce the DIMD mode, and these are called filterY and filterX, respectively. If the position where filterY and filterX are applied (i.e., the position of the reference sample) is E, the positions of the samples related to the application of the two filters can be represented as A, B, C, D, E, F, G, H, and I. If the values ​​obtained by applying filterY and filterX are represented as iDy and iDx, respectively, iDy and iDx can be calculated as follows. [Equation 6] Dy = A + 2*D + G - C - 2*F - I iDx = G + 2*H + I - A - 2*B - C If iDy and iDx are 0, the subsequent steps (i.e., obtaining the intensity value, selecting a specific intra prediction mode, and adding the intensity value to the accumulated intensity value for the intra prediction mode) can be skipped and the filter can be moved to the next sample position to be filtered. The function abs can be a function that calculates and returns the absolute value of an input value. The value of iAmp can be calculated as follows. This can correspond to step S710 of FIG. 7. [Equation 7] iAmp = abs(iDx) + abs(iDy) If either iDx or iDy is 0, iAngUneven, which is the value of the selected specific intra prediction mode, can be determined as in the following mathematical expression 8. This may correspond to step S720 of FIG. 7 for the case where either iDx or iDy is 0. [Equation 8] iAngUneven = ( iDx == 0 ) ? VER_IDX : HOR_IDX In Equation 8, VER_IDX and HOR_IDX can represent vertical mode and horizontal mode, respectively. For example, VER_IDX and HOR_IDX can correspond to mode 50 and mode 18 in Fig. 5, respectively. A value of iDx of 0 can mean that the amount of variation in the vertical direction is 0 or almost none. In other words, it can mean that the same sample values ​​are considered to be present in the vertical direction, and thus prediction is good in the vertical direction. Conversely, a value of iDy of 0 (i.e., when the value of iDx is not 0) can mean that the amount of variation in the horizontal direction is 0 or almost none. In other words, it can mean that the same sample values ​​are considered to be present in the horizontal direction, and thus prediction is good in the horizontal direction. If both iDx and iDy are non-zero, the value of iAngUneven, which is the value of the specific intra prediction mode selected, can be determined as follows. This may correspond to step S720 of FIG. 7 for the case where both iDx and iDy are non-zero. First, the values ​​of the intra prediction modes (especially the values ​​of the directional modes) can be divided into the following four groups. The first group (region 0) can be composed of modes having a specific number of horizontal directions. For example, the first group can be composed of modes having a value less than or equal to the horizontal mode. In Fig. 5, the modes having a horizontal direction are modes 2 to 34 (however, mode 34 may not be included in the modes having a horizontal direction), and the value of the horizontal mode is 18 (i.e., the horizontal mode is mode 18). If the specific number is N, the first group can be composed of modes {18, 18-1, 18-2, ... , 18-(N-1)}. If N is 17, the first group can be composed of modes {18, 17, 16, ... , 2}. The second group (region 1) can be composed of modes with a specific number of horizontal orientations. For example, the second group can be composed of modes whose values ​​are greater than or equal to the value of the horizontal mode. If the specific number is N, the second group can be composed of modes {18, 18+1, 18+2, ..., 18+(N-1)}. If N is 17, the second group can be composed of modes {18, 19, 20, ..., 34}. The third group (region 2) can be composed of modes having a specific number of vertical orientations. For example, the third group can be composed of modes having a value less than or equal to the vertical mode. In Fig. 5, the modes having a vertical orientation are modes 34 to 66 (however, mode 34 may not be included in the modes having a vertical orientation), and the value of the vertical mode is 50 (i.e., the vertical mode is mode 50). If a specific number is N, the third group can be composed of modes {50, 50-1, 50-2, ..., 50-(N-1)}. If N is 17, the third group can be composed of modes {50, 49, 48, ..., 34}. The fourth group (region 3) can be composed of modes with a specific number of vertical orientations. For example, the fourth group can be composed of modes whose values ​​are greater than or equal to the vertical mode. If the specific number is N, the fourth group can be composed of modes {50, 50+1, 50+2, ..., 50+(N-1)}. If N is 17, the fourth group can be composed of modes {50, 51, 52, ..., 66}. When defining groups for intra prediction modes as above, an identifier indicating a specific group can be calculated as shown in Table 14 below. The identifiers corresponding to the first to fourth groups are 0, 1, 2, and 3, respectively. signx = iDx < 0 ? 1 : 0signy = iDy < 0 ? 1 : 0absx = iDx < 0 ? -iDx : iDxabsy = iDy < 0 ? -iDy : iDygtY = absx > absy ? 1: 0mapXgrY1[0][0] = 1, mapXgrY1[0][1] = 0, mapXgrY1[1][0] = 0, mapXgrY1[1][1] = 1mapXgrY0[0][0] = 2, mapXgrY0[0][1] = 3, mapXgrY0[1][0] = 3, mapXgrY0[1][1] = 2region = gtY ? mapXgrY1[signy][signx] : mapXgrY0[signy][signx] In Table 14, gtY can indicate whether the amount of change in the vertical direction is greater than the amount of change in the horizontal direction. That is, if absx is greater than absy, gtY can be derived as 1, otherwise gtY can be derived as 0. Here, if gtY is 1, this indicates that the amount of change in the vertical direction is greater than the amount of change in the horizontal direction, and if gtY is 0, this indicates that the amount of change in the vertical direction is less than or equal to the amount of change in the horizontal direction. If the amount of change in the vertical direction is large, it may mean that the prediction in the vertical direction is less likely to be good. In this case, region 0 or region 1 can be selected. That is, mapXgrY1[signy][signx] can be selected as the identifier for the region. If the amount of change in the horizontal direction is large, it may mean that the prediction in the horizontal direction is less likely to be good. In this case, region 2 or region 3 can be selected. That is, mapXgrY0[signy][signx] can be selected as an identifier for the region. Also, for the horizontal direction, the direction that a positive change amount points to can be right, and the direction that a negative change amount points to can be left. For the vertical direction, the direction that a positive change amount points to can be downward, and the direction that a negative change amount points to can be upward. When the sign for iDx and the sign for iDy are the same, region 1 or region 2 can be selected, and when the sign for iDx and the sign for iDy are different, region 0 or region 3 can be selected. Next, the ratio value scaled to an integer value can be obtained as shown in Table 15 below. fRatio = gtY ? ( absy / absy ) : ( absx / absy )fRatioScaled = fRatio * (1 << 16)ratio = intFunc(fRatioScaled) In Table 15, (1 << 16) means shifting 1 to the left by 16, which is 2 16 represents 65536. The intFunc function can be a function that converts fRatioScaled expressed as a decimal value to an integer value. Operations such as rounding, ceiling, and flooring can be applied to convert to an integer value. You can also convert to an integer value using casting functions such as int provided by the C / C++ library. In Table 15, a division operation ( / ) may be required to obtain the ratio of absy and absx. However, when implementing a codec (especially when implementing in hardware), there is a problem that the implementation cost for the division operation is expensive or difficult to implement. Therefore, it is often advantageous in terms of implementation to approximate the division operation as a combination of several integer operations. Therefore, the formulas for obtaining the above ratio can be approximated by the formulas in Table 16 below. int g_gradDivTable

[0016] = { 0, 7, 6, 5, 5, 4, 4, 3, 3, 2, 2, 1, 1, 1, 1, 0};s0 = gtY ? absy : absxs1 = gtY ? absx : absyx = floorLog2(s1)norm = ((s1 << 4) >> x) & 15int v = g_gradDivTable[norm] | 8x = x + (norm != 0)shift = 13 - xif (shift < 0){shift = -shiftadd = (1 << (shift - 1))ratio = (s0 * v + add) >> shift}else{ratio = (s0 * v) << shift} The integerized ratio value can be determined through the process shown in Table 16. The position to which the ratio value is closest in angTable can be determined as shown in Table 17 below. int angTable

[0017] = { 0, 2048, 4096, 6144, 8192, 12288, 16384, 20480, 24576, 28672, 32768, 36864, 40960, 47104, 53248, 59392, 65536}idx = 16;for( int i = 1; i < 17; i++ ){if( ratio <= angTable[i] ){idx = ratio - angTable[i - 1] < angTable[i] - ratio ? i - 1 : i;break;}} According to Table 17, angTable can be composed of 17 entries. If each group described above consists of 17 modes, each intra-prediction mode constituting each group can correspond to an entry. In Table 17, the angTable entry closest to the ratio value can be found, and the value of idx can be derived based on the index of angleTable corresponding to that entry. int offsets[4] = { HOR_IDX, HOR_IDX, VER_IDX, VER_IDX}int dirs[4] = { -1, 1, -1, 1}iAngUneven = offsets[region] + dirs[region] * idx Table 18 is described according to the C / C++ syntax. iAngUneven can be assigned the value of the intra prediction mode for accumulating the intensity value for the current sample position within the current block to which the filter is applied. The region derived above can be an identifier indicating any one of the four groups. offsets[region] can indicate the starting value of the intra prediction mode for the group indicated by the region, and dirs[region] can indicate the direction in which the intra prediction mode increases or decreases. idx can indicate the position within the angleTable. Therefore, the value of iAngUneven can be determined using a formula such as Table 18. The process of adding the accumulated intensity values ​​for a selected specific intra prediction mode can be performed as follows. [Equation 9] piHistogram[iAngUneven] += iAmp In mathematical expression 9, the piHistogram array is an array that stores the accumulated intensity values ​​for all intra prediction modes, and after being initialized to 0, the intensity values ​​can be calculated by looping over the sample positions of the internal area within the current block to which the filter is applied. At this time, the intensity value calculated for the current sample position is added to the accumulated intensity value for the selected specific intra prediction mode. As described above, iAngUneven may store the value of the intra prediction mode selected for the current sample position to which the filter is applied, and iAmp may store the intensity value calculated for the corresponding sample position. piHistogram[iAngUneven] may store the accumulated intensity value for the intra prediction mode indicated by iAngUneven. The mathematical expression 9 may correspond to step S730 of FIG. 7. After looping over all sample locations within the internal region of the block to which the filter is applied, the piHistogram array stores the accumulated intensity values ​​for all intra prediction modes. One or more modes with the largest accumulated intensity values ​​can be selected. The selected mode may be referred to as the DIMD mode. This may correspond to step S740 in FIG. 7. firstAmp = 0, curAmp = 0;firstMode = 0, curMode = 0;for (int i = 0; i < NUM_LUMA_MODE; i++){curAmp = piHistogram[i];curMode = i;if (curAmp > firstAmp){firstAmp = curAmp;firstMode = curMode;}} Table 19 is described according to the C / C++ grammar. NUM_LUMA_MODE can represent the number of all intra prediction modes available in DIMD mode. For example, NUM_LUMA_MODE can be 67. According to Table 19, the value of the intra prediction mode with the largest accumulated intensity value can be assigned to the firstMode variable. Therefore, the intra prediction mode corresponding to the value of the finally determined firstMode variable can be set as the DIMD mode for the WxH block. Method 5 The DIMD of the aforementioned method 4 can be applied to the prediction block of the current block to derive an intra prediction mode, and the derived intra prediction mode can be set as a virtual intra prediction mode for the current block. The DIMD of the method 4 is applied to reconstructed samples belonging to the surrounding area of ​​the current block, and the DIMD of the method 5 is applied to prediction samples belonging to the prediction block of the current block, not the surrounding area of ​​the current block, so that the two are different, but the application method is otherwise the same. Specifically, the method 4 describes a method of deriving the DIMD mode by applying a PxQ filter (e.g., 3x3 filter) to a WxH block, and here, the WxH block can be set as a prediction block for the current block and the intra prediction mode can be derived by applying the DIMD. The prediction block of the current block may be generated based on at least one of intra prediction and inter prediction, as discussed in S410. Method 6 A virtual intra prediction mode for the current block can be derived based on template-based intra mode derivation (TIMD). TIMD can be a method for decoding devices to derive an intra prediction mode based on template regions adjacent to the current block. The intra prediction mode derived based on TIMD (hereinafter referred to as a TIMD mode) can be set as the virtual intra prediction mode for the current block. Below, the method for deriving the TIMD mode will be examined in detail. A cost can be calculated for each of the given candidate modes. The above-described candidate modes may refer to multiple intra prediction modes that are identically predefined for the encoding device and the decoding device. Alternatively, for template region-based derivation, a candidate list composed of the candidate modes may be generated, and costs may be calculated for candidate modes belonging to the candidate list. Alternatively, costs may be calculated only for the top N candidate modes within the generated candidate list. Here, N may be a value identically predefined for the encoding device and the decoding device. For example, N may be an integer of 2, 3, 4, 5, or higher. The above cost can be calculated as the sum of absolute differences (SAD) between the predicted samples and the reconstructed samples within the template region. Alternatively, the cost can also be calculated as the sum of absolute transformed differences (SATD) between the predicted samples and the reconstructed samples within the template region. Here, SATD may refer to the SAD transformed into the frequency domain. As an example of the transformation, the Hadamard transform may be used, but is not limited thereto. The predicted samples within the template region may be generated based on the candidate modes described above. The template area for calculating the above cost may be a pre-restored area adjacent to the current block. For example, the template area may include at least one of the upper peripheral area, the left peripheral area, the upper left peripheral area, the lower left peripheral area, or the upper right peripheral area of ​​the current block. Among the costs calculated for the above candidate modes, one candidate mode with the minimum cost may be selected. For example, costs may be calculated for each of five candidate modes in the candidate list. The five candidate modes in the candidate list may be reordered in ascending order of the calculated costs. The top candidate mode among the five reordered candidate modes may be selected. The selected candidate mode may be set as the TIMD mode. Alternatively, at least two candidate modes having the minimum cost among the costs calculated for the above candidate modes may be selected. For example, costs for five candidate modes in the candidate list may be calculated respectively. The five candidate modes in the candidate list may be reordered in ascending order of the calculated costs. The top two candidate modes among the reordered five candidate modes may be selected. When at least two candidate modes are selected through the above process, an intra prediction mode may be derived based on a comparison between the selected candidate modes and / or a comparison between at least one of the selected candidate modes and a threshold value. A TIMD mode may be derived based on the derived intra prediction mode. For example, an intra prediction mode can be derived based on whether the selected candidate modes satisfy the following conditions. [Condition] costMode2 < (K x costMode1) In the above condition, costMode1 may refer to a cost calculated based on one of the selected candidate modes, and costMode2 may refer to a cost calculated based on another of the selected candidate modes. For example, costMode1 may refer to a cost calculated based on a candidate mode with a smaller cost among the selected candidate modes, and costMode2 may refer to a cost calculated based on a candidate mode with a larger cost among the selected candidate modes. In the above condition, K represents a predetermined comparison factor, which may be a value that is identically predefined for the encoding device and the decoding device. For example, K may be an integer of 1, 2, or more, or may refer to a real number such as 1 / 2 or 1 / 4. If the above conditions are satisfied, an intra prediction mode can be derived based on the selected candidate modes. On the other hand, if the above conditions are not satisfied, a candidate mode with a cost of costMode1 may be derived as an intra prediction mode, and a candidate mode with a cost of costMode2 may not be used as an intra prediction mode. Method 7 If either of the two intra prediction modes derived by TIMD is used as the intra prediction mode of a partition within the current block, the corresponding intra prediction mode may be set as a virtual intra prediction mode for the current block. Alternatively, if the intra prediction modes of the two partitions are each identical to the two intra prediction modes derived by TIMD, either of the two partitions may be selected, and the intra prediction mode applied to the selected partition may be set as the virtual intra prediction mode. The method described in Method 3 may be applied as a method for selecting either of the two partitions. The two intra prediction modes derived by TIMD may be referred to as a horizontal TIMD mode and a vertical TIMD mode. Here, the horizontal TIMD mode may be a TIMD mode derived using the left peripheral region as a template region. The vertical TIMD mode may be a TIMD mode derived using the top peripheral region as a template region. The intra prediction mode for intra prediction of the current block may also be derived based on the aforementioned DIMD or TIMD, in which case the information for specifying the intra prediction modes of the two partitions within the current block (i.e., intra_pred_mode0_idx and intra_pred_mode1_idx) may not be signaled. A virtual intra prediction mode for determining a transformation set for non-separable transformation or an MTS set for separable transformation can be derived by sequentially checking one or more conditions. If a given condition is true, a virtual intra prediction mode can be derived based on a predefined method. On the other hand, if the given condition is false, a virtual intra prediction mode may not be derived or may be derived based on another predefined method. Here, one or more conditions may include at least one of conditions 1 to 6 described below, and the predefined method may be any one of methods 1 to 7 described above. [Condition 1] The intra prediction mode corresponding to the splitting direction of the current block is identical to at least one of the intra prediction modes applied to the two partitions within the current block. Here, the intra prediction mode corresponding to the splitting direction of the current block can be determined based on the aforementioned method 1. [Condition 2] The distance between the dividing line of the current block and the center of the current block is greater than or equal to a predetermined distance (d). Here, the predetermined distance (d) may mean an actual distance, or a division type of the current block or a corresponding distance index. As the dividing line moves away from the center of the current block, one partition becomes larger than another partition. In this case, the intra prediction mode for the larger partition may be considered an intra prediction mode that better represents the entire current block. When the value of the predetermined distance (d) is defined by the value of the distance index (distanceIdx) mentioned above, the value of d may be 0, 1, 2, or 3. The fact that the dividing line of the current block is d or more away from the center of the current block may mean that the value of distanceIdx is d or more. [Condition 3] The intra prediction mode corresponding to the split direction of the current block is identical to the intra prediction mode applied to the larger partition among the two partitions within the current block. Here, the intra prediction mode corresponding to the split direction of the current block can be determined based on the aforementioned method 1. [Condition 4] The distance between the division line of the current block and the center of the current block is greater than or equal to a predetermined distance (d), and the intra prediction mode corresponding to the division direction of the current block is the same as the intra prediction mode applied to the larger partition among the two partitions within the current block. The predetermined distance (d) is as discussed in Condition 2. Here, the intra prediction mode corresponding to the division direction of the current block can be determined based on the aforementioned Method 1. [Condition 5] The distance between the dividing line of the current block and the center of the current block is 0 (or, the distanceIdx value is 0), and the intra prediction mode corresponding to the dividing direction of the current block is identical to at least one of the intra prediction modes applied to the two partitions within the current block. If the distance between the dividing line of the current block and the center of the current block is 0, this may mean that the sizes of the two partitions are the same. Here, the intra prediction mode corresponding to the dividing direction of the current block may be determined based on the aforementioned method 1. [Condition 6] The distance between the division line of the current block and the center of the current block is greater than 0 (or, the distanceIdx value is greater than 0), and the intra prediction mode corresponding to the division direction of the current block is the same as the intra prediction mode applied to the larger partition among the two partitions within the current block. Here, the intra prediction mode corresponding to the division direction of the current block can be determined based on the aforementioned method 1. For example, if the intra prediction mode corresponding to the partitioning direction of the current block is identical to at least one of the intra prediction modes applied to two partitions within the current block (if Condition 1 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 1 is false), the virtual intra prediction mode for the current block can be derived based on Method 3. Alternatively, if the intra prediction mode corresponding to the splitting direction of the current block is identical to at least one of the intra prediction modes applied to two partitions within the current block (if Condition 1 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 1 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the distance between the dividing line of the current block and the center of the current block is greater than or equal to 3 (condition 2 where d is 3 is true), the virtual intra prediction mode for the current block can be derived based on method 3. Otherwise (condition 2 where d is 3 is false), the virtual intra prediction mode for the current block can be derived based on method 1. Alternatively, if the distance between the dividing line of the current block and the center of the current block is greater than or equal to 2 (condition 2 where d is 2 is true), the virtual intra prediction mode for the current block can be derived based on method 3. Otherwise (condition 2 where d is 2 is false), the virtual intra prediction mode for the current block can be derived based on method 1. Alternatively, if the distance between the dividing line of the current block and the center of the current block is greater than or equal to 1 (condition 2 where d is 1 is true), the virtual intra prediction mode for the current block can be derived based on method 3. Otherwise (condition 2 where d is 1 is false), the virtual intra prediction mode for the current block can be derived based on method 1. Alternatively, if the distance between the dividing line of the current block and the center of the current block is greater than or equal to 3 (condition 2 where d is 3 is true), the virtual intra prediction mode for the current block can be derived based on method 3. Otherwise (condition 2 where d is 3 is false), the virtual intra prediction mode for the current block can be derived based on method 5. Alternatively, if the distance between the dividing line of the current block and the center of the current block is greater than or equal to 2 (condition 2 where d is 2 is true), the virtual intra prediction mode for the current block can be derived based on method 3. Otherwise (condition 2 where d is 2 is false), the virtual intra prediction mode for the current block can be derived based on method 5. Alternatively, if the distance between the dividing line of the current block and the center of the current block is greater than or equal to 1 (condition 2 where d is 1 is true), the virtual intra prediction mode for the current block can be derived based on method 3. Otherwise (condition 2 where d is 1 is false), the virtual intra prediction mode for the current block can be derived based on method 5. Alternatively, if the distance between the dividing line of the current block and the center of the current block is greater than or equal to 3 (if Condition 2 is true that d is 3), the virtual intra prediction mode for the current block can be derived based on Method 3. Otherwise (if Condition 2 is false that d is 3), it can be checked whether the intra prediction mode corresponding to the dividing direction of the current block is identical to at least one of the intra prediction modes applied to the two partitions within the current block. If the intra prediction mode corresponding to the dividing direction of the current block is identical to at least one of the intra prediction modes applied to the two partitions within the current block (if Condition 1 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 1 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the distance between the dividing line of the current block and the center of the current block is greater than or equal to 2 (if Condition 2 where d is 2 is true), the virtual intra prediction mode for the current block can be derived based on Method 3. Otherwise (if Condition 2 where d is 2 is false), it can be checked whether the intra prediction mode corresponding to the dividing direction of the current block is identical to at least one of the intra prediction modes applied to the two partitions within the current block. If the intra prediction mode corresponding to the dividing direction of the current block is identical to at least one of the intra prediction modes applied to the two partitions within the current block (if Condition 1 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 1 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the distance between the splitting line of the current block and the center of the current block is greater than or equal to 1 (if Condition 2 is true that d is 1), the virtual intra prediction mode for the current block can be derived based on Method 3. Otherwise (if Condition 2 is false that d is 1), it can be checked whether the intra prediction mode corresponding to the splitting direction of the current block is identical to at least one of the intra prediction modes applied to the two partitions within the current block. If the intra prediction mode corresponding to the splitting direction of the current block is identical to at least one of the intra prediction modes applied to the two partitions within the current block (if Condition 1 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 1 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the distance between the splitting line of the current block and the center of the current block is greater than or equal to 3 (if Condition 2 is true that d is 3), the virtual intra prediction mode for the current block can be derived based on Method 3. Otherwise (if Condition 2 is false that d is 3), it can be checked whether the intra prediction mode corresponding to the splitting direction of the current block is the same as the intra prediction mode applied to the larger partition among the two partitions in the current block. If the intra prediction mode corresponding to the splitting direction of the current block is the same as the intra prediction mode applied to the larger partition among the two partitions in the current block (if Condition 3 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 3 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the distance between the splitting line of the current block and the center of the current block is greater than or equal to 2 (if Condition 2 where d is 2 is true), the virtual intra prediction mode for the current block can be derived based on Method 3. Otherwise (if Condition 2 where d is 2 is false), it can be checked whether the intra prediction mode corresponding to the splitting direction of the current block is the same as the intra prediction mode applied to the larger partition among the two partitions in the current block. If the intra prediction mode corresponding to the splitting direction of the current block is the same as the intra prediction mode applied to the larger partition among the two partitions in the current block (if Condition 3 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 3 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the distance between the splitting line of the current block and the center of the current block is greater than or equal to 1 (if Condition 2 is true that d is 1), the virtual intra prediction mode for the current block can be derived based on Method 3. Otherwise (if Condition 2 is false that d is 1), it can be checked whether the intra prediction mode corresponding to the splitting direction of the current block is identical to the intra prediction mode applied to the larger partition among the two partitions in the current block. If the intra prediction mode corresponding to the splitting direction of the current block is identical to the intra prediction mode applied to the larger partition among the two partitions in the current block (if Condition 3 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 3 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the intra prediction mode corresponding to the splitting direction of the current block is the same as at least one of the intra prediction modes applied to two partitions within the current block (if Condition 1 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 1 is false), it can be checked whether the distance between the splitting line of the current block and the center of the current block is greater than or equal to 3. If the distance between the splitting line of the current block and the center of the current block is greater than or equal to 3 (if Condition 2 that d is 3 is true), the virtual intra prediction mode for the current block can be derived based on Method 3. Otherwise (if Condition 2 that d is 3 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the intra prediction mode corresponding to the splitting direction of the current block is the same as at least one of the intra prediction modes applied to two partitions within the current block (if Condition 1 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 1 is false), it can be checked whether the distance between the splitting line of the current block and the center of the current block is greater than or equal to 2. If the distance between the splitting line of the current block and the center of the current block is greater than or equal to 2 (if Condition 2 with d being 2 is true), the virtual intra prediction mode for the current block can be derived based on Method 3. Otherwise (if Condition 2 with d being 2 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the intra prediction mode corresponding to the splitting direction of the current block is the same as at least one of the intra prediction modes applied to two partitions within the current block (if Condition 1 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 1 is false), it can be checked whether the distance between the splitting line of the current block and the center of the current block is greater than or equal to 1. If the distance between the splitting line of the current block and the center of the current block is greater than or equal to 1 (if Condition 2 that d is 1 is true), the virtual intra prediction mode for the current block can be derived based on Method 3. Otherwise (if Condition 2 that d is 1 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the intra prediction mode corresponding to the splitting direction of the current block is the same as the intra prediction mode applied to the larger partition among the two partitions in the current block (if Condition 3 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 3 is false), it can be checked whether the distance between the splitting line of the current block and the center of the current block is greater than or equal to 3. If the distance between the splitting line of the current block and the center of the current block is greater than or equal to 3 (if Condition 2 where d is 3 is true), the virtual intra prediction mode for the current block can be derived based on Method 3. Otherwise (if Condition 2 where d is 3 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the intra prediction mode corresponding to the splitting direction of the current block is the same as the intra prediction mode applied to the larger partition among the two partitions in the current block (if Condition 3 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 3 is false), it can be checked whether the distance between the splitting line of the current block and the center of the current block is greater than or equal to 2. If the distance between the splitting line of the current block and the center of the current block is greater than or equal to 2 (if Condition 2 with d being 2 is true), the virtual intra prediction mode for the current block can be derived based on Method 3. Otherwise (if Condition 2 with d being 2 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the intra prediction mode corresponding to the splitting direction of the current block is the same as the intra prediction mode applied to the larger partition of the two partitions in the current block (if Condition 3 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 3 is false), it can be checked whether the distance between the splitting line of the current block and the center of the current block is greater than or equal to 1. If the distance between the splitting line of the current block and the center of the current block is greater than or equal to 1 (if Condition 2 that d is 1 is true), the virtual intra prediction mode for the current block can be derived based on Method 3. Otherwise (if Condition 2 that d is 1 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the intra prediction mode corresponding to the splitting direction of the current block is the same as the intra prediction mode applied to the larger partition among the two partitions in the current block (if Condition 3 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 3 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the distance between the dividing line of the current block and the center of the current block is greater than or equal to 3, and the intra prediction mode corresponding to the dividing direction of the current block is the same as the intra prediction mode applied to the larger partition among the two partitions in the current block (if Condition 4 with d being 3 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 4 with d being 3 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the distance between the dividing line of the current block and the center of the current block is greater than or equal to 2, and the intra prediction mode corresponding to the dividing direction of the current block is the same as the intra prediction mode applied to the larger partition among the two partitions in the current block (if Condition 4 with d being 2 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 4 with d being 2 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the distance between the dividing line of the current block and the center of the current block is greater than or equal to 1, and the intra prediction mode corresponding to the dividing direction of the current block is the same as the intra prediction mode applied to the larger partition among the two partitions in the current block (if Condition 4 with d being 1 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 4 with d being 1 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the distance between the splitting line of the current block and the center of the current block is 0 and the intra prediction mode corresponding to the splitting direction of the current block is equal to at least one of the intra prediction modes applied to the two partitions within the current block (if Condition 5 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 5 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Alternatively, if the distance between the dividing line of the current block and the center of the current block is greater than 0 and the intra prediction mode corresponding to the dividing direction of the current block is the same as the intra prediction mode applied to the larger partition among the two partitions in the current block (if Condition 6 is true), the virtual intra prediction mode for the current block can be derived based on Method 1. Otherwise (if Condition 6 is false), the virtual intra prediction mode for the current block can be derived based on Method 5. Some existing codecs employ Multiple Transform Selection (MTS) technology. Adaptive MTS technology can also be considered part of MTS. When geometric partitioning is applied, other MTS methods can be applied. For example, implicit MTS, rather than explicit MTS, can be applied when geometric partitioning is applied. Implicit MTS refers to a method that implicitly determines the horizontal and vertical transformation kernels based on the size of the current block. In this case, the signaling of an index to specify one of the transformation kernel candidates can be omitted. Specifically, if the width of the current block is 4 or more and 16 or less, DST-7 can be applied as the horizontal transformation. Otherwise, DCT-2 can be applied as the horizontal transformation. If the height of the current block is 4 or more and 16 or less, DST-7 can be applied as the vertical transformation. Otherwise, DCT-2 can be applied as the vertical transformation. Alternatively, when geometric partitioning is applied, explicit MTS can be applied instead of adaptive MTS. Here, explicit MTS refers to a method of selectively applying any one of DCT-2 pairs (i.e., (DCT-2, DCT-2)) and combinations of DST-7 and DCT-8 (i.e., (DST-7, DST-7), (DCT-8, DST-7), (DST-7, DCT-8), (DCT-8, DCT-8)). For this purpose, an index can be signaled to specify any one of the transform kernel candidates. Referring to FIG. 4, the current block can be restored based on the prediction block and residual block of the current block (S430). FIG. 9 illustrates a schematic configuration of a decoding device (300) that performs an image decoding method according to the present disclosure. Referring to FIG. 9, a decoding device (300) according to the present disclosure may include a block division unit (900), a prediction block generation unit (910), a residual block derivation unit (920), and a restoration block generation unit (930). The block division unit (900) and the prediction block generation unit (910) may be configured in the prediction unit (330) of FIG. 3, the residual block derivation unit (920) may be configured in the residual processing unit (320) of FIG. 3, and the restoration block generation unit (930) may be configured in the addition unit (340) of FIG. 3. The block partitioning unit (900) can partition a current block into multiple partitions. The current block may be partitioned based on one or more partitioning lines, or geometric partitioning may be applied to the current block. The prediction block generation unit (910) can generate a prediction block of the current block by performing intra-prediction or inter-prediction on each partition within the current block. The method for generating the prediction block is as described with reference to FIG. 4, and any redundant description will be omitted here. The residual block derivation unit (920) can derive the residual block of the current block based on the transform coefficient of the current block. The entropy decoding unit (310) can derive the transform coefficient of the current block based on residual information signaled through the bitstream. The residual block derivation unit (920) can derive the residual block by performing at least one of dequantization and inverse transform on the transform coefficient of the current block. In addition, the residual block derivation unit (920) can determine a transformation kernel for the inverse transformation of the current block through a predetermined transformation kernel determination method, and derive the residual block of the current block based on this. This has been described with reference to FIG. 4, and any duplicate description will be omitted here. The restoration block generation unit (930) can restore the current block based on the prediction block and residual block of the current block. FIG. 10 illustrates an image encoding method performed by an encoding device (200) as an embodiment according to the present disclosure. Referring to Fig. 10, the current block can be divided into multiple partitions (S1000). This is as discussed with reference to Fig. 4. Referring to FIG. 10, a prediction block of the current block can be generated based on the prediction for each partition (S1010). As discussed with reference to Fig. 4, when a current block is divided into two partitions (i.e., a first partition and a second partition), the prediction block of the current block can be generated as a weighted sum of a first prediction block for the first partition and a second prediction block for the second partition. Here, each of the first and second prediction blocks can be generated based on intra prediction or inter prediction. When the aforementioned SGPM is applied to the current block, the partition type of the current block and the intra prediction mode of each partition can be determined. A partition type index for indicating the determined partition type can be encoded in a bitstream. A mode index indicating the determined intra prediction mode among a plurality of intra prediction mode candidates can be encoded in the bitstream. The mode index can be encoded for each partition. The mode index for the second partition can be encoded based on the mode index for the first partition. At least one of the partition type index or the mode index can be encoded based on a flag (cu_sgpm_flag) indicating whether SGPM is applied to the current block. As described with reference to FIG. 4, a candidate list for the SGPM of the current block may be constructed. Each candidate in the candidate list may include one partition type index and two mode indices. Based on any one of the multiple candidates in the candidate list, a partition type index for the current block and a mode index for each partition may be derived. A candidate index indicating any one of the multiple candidates may be encoded in the bitstream. In addition, the candidate list may be reordered based on a predetermined template region. As discussed with reference to Figure 4, the intra prediction mode candidates can be organized into an intra prediction mode (IPM) list. SGPM can also be applied when the current block size satisfies certain conditions. It may be determined whether blending between the prediction blocks of the first partition and the prediction blocks of the second partition for the current block is allowed. If it is determined that blending between the prediction blocks is allowed, adaptive blending may be used in the SGPM, and the blending depth for the adaptive blending may be derived based on the size of the current block. On the other hand, if it is determined that blending between the prediction blocks is not allowed, the blending depth may be derived as a default value (e.g., 1 / 4τ). Based on the determination, a flag indicating whether blending between the prediction blocks of the first and second partitions is allowed may be encoded in the bitstream. Referring to FIG. 10, a residual block of the current block can be derived based on the predicted block of the current block (S1020). The residual block of the current block can be derived by subtracting the predicted block from the original block of the current block. Referring to FIG. 10, transform coefficients of the current block can be derived from the residual block of the current block (S1030). Transform coefficients can be derived by performing at least one of transform and quantization on the residual block of the current block. The transformation according to the present disclosure can be understood as the reverse process of the inverse transformation discussed with reference to FIG. 4. The method for determining the transformation kernel for the above transformation is as discussed with reference to FIG. 4, and a detailed description thereof will be omitted here. Referring to FIG. 10, a bitstream can be generated by encoding the transform coefficients of the current block (S1040). Based on the transform coefficients of the current block, residual information regarding the transform coefficients can be generated, and a bitstream can be generated by encoding the residual information. FIG. 11 illustrates a schematic configuration of an encoding device (200) that performs an image encoding method according to the present disclosure. Referring to FIG. 11, an encoding device (200) according to the present disclosure may include a block division unit (1100), a prediction block generation unit (1110), a residual block derivation unit (1120), a transform coefficient derivation unit (1130), and a transform coefficient encoding unit (1140). The block division unit (1100) and the prediction block generation unit (1110) may be configured in the prediction unit (220) of FIG. 2, the residual block derivation unit (1120) and the transform coefficient derivation unit (1130) may be configured in the residual processing unit (230) of FIG. 2, and the transform coefficient encoding unit (1130) may be configured in the entropy encoding unit (240) of FIG. 2. The block partitioning unit (1100) can partition the current block into multiple partitions. That is, the current block can be partitioned based on one or more partitioning lines, or geometric partitioning can be applied to the current block. The prediction block generation unit (1110) can generate a prediction block of the current block by performing intra-prediction or inter-prediction on each partition within the current block. The method for generating the prediction block is as described with reference to FIG. 4, and any redundant description will be omitted here. The residual block derivation unit (1120) can derive a residual block of the current block based on the predicted block of the current block. The transform coefficient derivation unit (1130) can derive the transform coefficients of the current block from the residual block of the current block. That is, the transform coefficients can be derived by performing at least one of transform and quantization on the residual block of the current block. Here, the transform can be understood as the reverse process of the inverse transform discussed with reference to FIG. 4. The method for determining the transform kernel for the transform is the same as discussed with reference to FIG. 4, and a detailed description thereof will be omitted here. The transform coefficient encoding unit (1140) can encode the transform coefficient of the current block to generate a bitstream. Based on the transform coefficient of the current block, residual information regarding the transform coefficient can be generated, and a bitstream can be generated by encoding the residual information. In the embodiments described above, the methods are described based on a flowchart as a series of steps or blocks. However, the embodiments are not limited to the order of the steps, and some steps may occur in a different order or simultaneously with other steps described above. Furthermore, those skilled in the art will understand that the steps depicted in the flowchart are not exclusive, and other steps may be included, or one or more steps in the flowchart may be deleted without affecting the scope of the embodiments of this document. The method according to the embodiments of the present document described above can be implemented in the form of software, and the encoding device and / or decoding device according to the present document can be included in a device that performs image processing, such as a TV, a computer, a smartphone, a set-top box, a display device, etc. When the embodiments in this document are implemented as software, the above-described method can be implemented as a module (process, function, etc.) that performs the above-described function. The module can be stored in memory and executed by a processor. The memory can be internal or external to the processor and can be connected to the processor by various well-known means. The processor can include an application-specific integrated circuit (ASIC), another chipset, logic circuit, and / or data processing device. The memory can include a read-only memory (ROM), a random access memory (RAM), flash memory, a memory card, a storage medium, and / or other storage devices. That is, the embodiments described in this document can be implemented and performed on a processor, a microprocessor, a controller, or a chip. For example, the functional units illustrated in each drawing can be implemented and performed on a computer, a processor, a microprocessor, a controller, or a chip. In this case, information for implementation (e.g., information on instructions) or an algorithm can be stored on a digital storage medium. In addition, the decoding device and encoding device to which the embodiment(s) of the present specification are applied may be included in a multimedia broadcasting transmitting and receiving device, a mobile communication terminal, a home cinema video device, a digital cinema video device, a surveillance camera, a video conversation device, a real-time communication device such as a video communication, a mobile streaming device, a storage medium, a camcorder, a video-on-demand (VoD) service providing device, an OTT (Over the top video) device, an Internet streaming service providing device, a three-dimensional (3D) video device, a VR (virtual reality) device, an AR (argumente reality) device, a video phone video device, a transportation terminal (ex. a vehicle (including an autonomous vehicle) terminal, an airplane terminal, a ship terminal, etc.), and a medical video device, and may be used to process a video signal or a data signal. For example, the OTT (Over the top video) device may include a game console, a Blu-ray player, an Internet-connected TV, a home theater system, a smartphone, a tablet PC, a DVR (Digital Video Recorder), etc. In addition, the processing method to which the embodiment(s) of the present specification are applied can be produced in the form of a computer-executable program and can be stored in a computer-readable recording medium. Multimedia data having a data structure according to the embodiment(s) of the present specification can also be stored in a computer-readable recording medium. The computer-readable recording medium includes all types of storage devices and distributed storage devices in which computer-readable data is stored. The computer-readable recording medium can include, for example, a Blu-ray disc (BD), a universal serial bus (USB), a ROM, a PROM, an EPROM, an EEPROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device. In addition, the computer-readable recording medium includes a medium implemented in the form of a carrier wave (e.g., transmission via the Internet). In addition, a bitstream generated by an encoding method can be stored in a computer-readable recording medium or transmitted via a wired or wireless communication network. Additionally, the embodiments of the present disclosure may be implemented as a computer program product by program code, and the program code may be executed on a computer by the embodiments of the present disclosure. The program code may be stored on a computer-readable carrier. FIG. 12 illustrates an example of a content streaming system to which embodiments of the present disclosure can be applied. Referring to FIG. 12, a content streaming system to which the embodiment(s) of the present specification are applied may largely include an encoding server, a streaming server, a web server, a media storage, a user device, and a multimedia input device. The encoding server compresses content input from multimedia input devices such as smartphones, cameras, and camcorders into digital data, generates a bitstream, and transmits it to the streaming server. Alternatively, if multimedia input devices such as smartphones, cameras, and camcorders directly generate bitstreams, the encoding server may be omitted. The above bitstream can be generated by an encoding method or a bitstream generation method to which the embodiment(s) of the present specification are applied, and the streaming server can temporarily store the bitstream during the process of transmitting or receiving the bitstream. The streaming server transmits multimedia data to a user device based on a user request via a web server, and the web server acts as an intermediary to inform the user of available services. When a user requests a desired service from the web server, the web server transmits the request to the streaming server, and the streaming server transmits the multimedia data to the user. At this time, the content streaming system may include a separate control server, in which case the control server controls commands / responses between each device within the content streaming system. The streaming server can receive content from a media repository and / or an encoding server. For example, when receiving content from the encoding server, the content can be received in real time. In this case, to provide a smooth streaming service, the streaming server can store the bitstream for a certain period of time. Examples of the user devices may include mobile phones, smart phones, laptop computers, digital broadcasting terminals, personal digital assistants (PDAs), portable multimedia players (PMPs), navigation devices, slate PCs, tablet PCs, ultrabooks, wearable devices (e.g., smartwatches, smart glasses, HMDs), digital TVs, desktop computers, digital signage, etc. Each server within the above content streaming system can be operated as a distributed server, in which case data received from each server can be processed in a distributed manner. The claims set forth in this specification may be combined in various ways. For example, the technical features of the method claims of this specification may be combined and implemented as a device, and the technical features of the device claims of this specification may be combined and implemented as a method. Furthermore, the technical features of the method claims and the technical features of the device claims of this specification may be combined and implemented as a device, and the technical features of the method claims and the technical features of the device claims of this specification may be combined and implemented as a method.

Claims

1. A step of dividing the current block into multiple partitions; A step of generating a prediction block of the current block by performing a prediction for each of the above partitions; A step of deriving a residual block of the current block based on an inverse transformation for the current block; and A step of restoring the current block based on the predicted block and residual block of the current block, The step of deriving the residual block of the current block above is: a step of deriving a virtual intra prediction mode for the current block; and A method comprising the step of determining a transform set for the inverse transformation based on the virtual intra prediction mode.

2. In paragraph 1, The above virtual intra prediction mode is a method in which a pre-defined fixed mode is derived.

3. In paragraph 1, A method in which the above virtual intra prediction mode is derived based on a division direction according to a geometric division of the current block.

4. In paragraph 3, A method in which the above virtual intra prediction mode is derived as an intra prediction mode that matches or is closest to a division direction according to a geometric division of the current block.

5. In paragraph 3, A method in which the above virtual intra prediction mode is derived as an intra prediction mode that matches or is closest to a direction perpendicular to a division direction according to a geometric division of the current block.

6. In paragraph 1, A method wherein the virtual intra prediction mode is derived based on at least one of a partition type of the current block or a partition index corresponding to the partition type.

7. In paragraph 1, The above virtual intra prediction mode is derived as a DIMD (decoder side intra mode derivation) mode, A method in which the above DIMD mode is derived based on restored samples belonging to a surrounding area of ​​the current block.

8. In paragraph 1, The above virtual intra prediction mode is derived as a DIMD (decoder side intra mode derivation) mode, A method in which the above DIMD mode is derived based on prediction samples belonging to a prediction block of the current block.

9. In paragraph 1, A method wherein the above virtual intra prediction mode is derived based on one or more TIMD (template-based intra mode derivation) modes for the current block.

10. In paragraph 1, A method in which the virtual intra prediction mode is derived based on at least one of whether the intra prediction mode corresponding to the division direction according to the geometric division of the current block is identical to at least one of the intra prediction modes applied to the plurality of partitions or the distance between the division line for the geometric division of the current block and the center of the current block.

11. Step of dividing the current block into multiple partitions; A step of generating a prediction block of the current block by performing a prediction for each of the above partitions; A step of deriving a residual block of the current block based on a predicted block of the current block; A step of deriving transform coefficients of the current block based on a transform of a residual block of the current block; and Including a step of encoding the transform coefficients of the current block, The step of deriving the transformation coefficients of the current block is: a step of deriving a virtual intra prediction mode for the current block; and A method comprising the step of determining a transform set for the transformation based on the virtual intra prediction mode.

12. A computer-readable storage medium storing a bitstream generated by the method according to Article 11.

13. A step of obtaining a bitstream for image information; wherein the bitstream is generated based on a step of dividing a current block into a plurality of partitions, a step of performing prediction on each of the partitions to generate a prediction block of the current block, a step of deriving a residual block of the current block based on the prediction block of the current block, a step of deriving transform coefficients of the current block based on a transform of the residual block of the current block, and a step of encoding transform coefficients of the current block, and Including a step of transmitting data including the above bitstream, Including a step of encoding the transform coefficients of the current block, The step of deriving the transformation coefficients of the current block is: a step of deriving a virtual intra prediction mode for the current block; and A method comprising the step of determining a transform set for the transformation based on the virtual intra prediction mode.

Images