Video coding method and apparatus using improved in-loop filter for chroma components

By using cross-component resampling in-loop filters (CC-RIF) in video coding, the problem of insufficient coding efficiency in existing technologies is solved, and more efficient video quality improvement is achieved, especially by effectively utilizing upsampling or super-resolution techniques in the in-loop filtering of the chroma component.

CN119032575APending Publication Date: 2024-11-26HYUNDAI MOTOR CO LTD +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202380034162.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2023-03-22
Filing Date
2023-03-23
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

Existing video coding technologies are inefficient when processing high-resolution and high-frame-rate videos, making it difficult to effectively utilize upsampling or super-resolution techniques to improve video quality.

Method used

The Cross-Component Resampling In-Loop Filter (CC-RIF) is used to generate upsampled chroma frames by inputting luminance and chroma samples into the CC-RIF filter. This replaces or inserts existing chroma component filters, including deblocking filters, sampling adaptive offset filters, and adaptive loop filters, thereby improving video coding efficiency.

Benefits of technology

It improves video coding efficiency and enhances video quality, especially by effectively utilizing upsampling or super-resolution techniques in the in-loop filtering of the chroma component, thereby improving image enhancement effects.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119032575A_ABST
    Figure CN119032575A_ABST
Patent Text Reader

Abstract

Disclosed are a video encoding method and apparatus using an improved in-loop filter for chroma components. In this embodiment, a video decoding device obtains a reconstructed frame including luma samples and chroma samples, and then obtains a scaled value representing a resolution difference between the reconstructed frame and an original frame. When the resolution of the reconstructed frame based on the scaled value is less than the resolution of the original frame, the video decoding device inputs the luma samples and the chroma samples into a cross-component resampling in-loop filter (CC-RIF) to generate an up-sampled chroma frame.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a video encoding method and apparatus using an improved in-loop filter for chrominance components. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and may not constitute prior art.

[0003] Since video data has a large amount of data compared to audio or still image data, without a process for compression, the video data requires a large amount of hardware resources (including memory) to store or transmit the video data.

[0004] Thus, encoders are generally used to compress and store or transmit video data. A decoder receives compressed video data, decompresses the received compressed video data, and plays the decompressed video data. Video compression technologies include H.264 / Advanced Video Coding (AVC), High Efficiency Video Coding (HEVC), and Versatile Video Coding (VVC), which have improved coding efficiency of about 30% or more compared to HEVC.

[0005] However, as image size, resolution and frame rate gradually increase, the amount of data to be encoded also increases. Therefore, new compression technologies that provide higher coding efficiency and improved image enhancement effects than existing compression technologies are needed. In particular, in the in-loop filtering of chrominance components, methods that effectively utilize upsampling or super-resolution techniques are needed to improve video coding efficiency and enhance video quality. Summary of the invention

[0006]

Technical issues

[0007] The present invention attempts to provide a video encoding method and apparatus for reconstructing chrominance components. In order to improve video encoding efficiency and enhance video quality, the video encoding method and apparatus utilize an in-loop filter based on upsampling or super-resolution technology.

[0008]

Technical solution

[0009] At least one aspect of the present invention provides a method for upsampling a reconstructed frame performed by a video decoding device. The method includes obtaining a reconstructed frame, which is a reconstruction of an original frame and includes a luminance sample and a chrominance sample. The method also includes obtaining a scaling value representing the resolution difference between the reconstructed frame and the original frame. The method also includes: based on the scaling value, generating an upsampled chrominance frame by inputting the luminance sample and the chrominance sample into a cross-component resampling in-loop filter (CC-RIF). Here, the CC-RIF replaces one of the filters for the chrominance component, the filters including a deblocking filter, a sampling adaptive offset filter (SAO filter), and an adaptive loop filter, or the CC-RIF is inserted between filters.

[0010] Another aspect of the present invention provides a method for upsampling a reconstructed frame performed by a video encoding device. The method includes obtaining a scaling value representing the resolution difference between the reconstructed frame and the original frame. The reconstructed frame is a reconstruction of the original frame and includes a luminance sample and a chrominance sample. The method also includes obtaining the reconstructed frame. The method includes: when the reconstructed frame has a resolution based on the scaling value and is less than the resolution of the original frame, an upsampled chrominance frame is generated by inputting the luminance sample and the chrominance sample into a cross-component resampling in-loop filter (CC-RIF). Here, the CC-RIF replaces one of the filters for the chrominance component, the filters including a deblocking filter, a sampling adaptive offset filter (SAO filter) and an adaptive loop filter, or the CC-RIF is inserted between filters.

[0011] Another aspect of the present invention provides a computer-readable recording medium storing a bitstream generated by a video encoding method. The video encoding method includes: obtaining a scaling value representing the resolution difference between a reconstructed frame and an original frame. The reconstructed frame is a reconstruction of the original frame and includes a luminance sample and a chrominance sample. The video encoding method also includes obtaining the reconstructed frame. The video encoding method also includes: when the reconstructed frame has a resolution based on the scaling value and is less than the resolution of the original frame, an upsampled chrominance frame is generated by inputting the luminance sample and the chrominance sample into a cross-component resampling in-loop filter (CC-RIF). Here, the CC-RIF replaces one of the filters for the chrominance component, the filter including a deblocking filter, a sampling adaptive offset filter (SAO filter) and an adaptive loop filter, or the CC-RIF is inserted between filters.

[0012]

Beneficial Effects

[0013] As described above, the present invention provides a video encoding method and apparatus that utilizes an in-loop filter based on upsampling or super-resolution technology to reconstruct chrominance components. Thus, the video encoding method and apparatus improve video encoding efficiency and enhance video quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a block diagram of a video encoding device that may implement the techniques of this disclosure.

[0015] Figure 2 A method for partitioning blocks using a quadtree plus binary tree ternary tree (QTBTTT) structure is described.

[0016] Figure 3a and 3b Multiple intra prediction modes are shown including a wide-angle intra prediction mode.

[0017] Figure 4 Neighboring blocks of the current block are shown.

[0018] Figure 5 is a block diagram of a video decoding device that may implement the techniques of this disclosure.

[0019] Figure 6 is a diagram illustrating the operation of a convolutional layer.

[0020] Figure 7 is a diagram illustrating the operation of a deconvolution layer.

[0021] Figure 8 is a diagram illustrating the operation of a pooling layer.

[0022] Fig. 9 is a diagram showing a single image super-resolution (SISR) network.

[0023] Fig.10 is a diagram showing a residual block utilized in SISR.

[0024] Fig.11 is a diagram showing a form of an adaptive loop filter (ALF).

[0025] Fig.12 is a diagram showing application of a cross-component adaptive loop filter (CC-ALF).

[0026] Fig.13 is a diagram showing the format of CC-ALF.

[0027] Fig.14 is a block diagram illustrating a video decoding apparatus with reference picture resampling (RPR) technology.

[0028] Fig.15is a diagram illustrating pixel shuffling according to at least one embodiment of the present disclosure.

[0029] Fig.16 is a diagram illustrating a chroma component in-loop filter (CC-IF) based on pixel shuffling according to at least one embodiment of the present disclosure.

[0030] Fig.17 is a diagram illustrating channel attention according to at least one embodiment of the present disclosure.

[0031] Fig.18 is a diagram illustrating a CC-IF according to at least one embodiment of the present disclosure.

[0032] Fig.19 is a diagram illustrating a video decoding apparatus including a chroma component resampling in-loop filter (C-RIF) according to at least one embodiment of the present disclosure.

[0033] Fig. 20 is a diagram illustrating a transposed convolution operation.

[0034] Fig.21 is a diagram illustrating application of a cross-component resampling in-loop filter (CC-RIF) in accordance with at least one embodiment of the present disclosure.

[0035] Fig. 22 is a diagram illustrating the application of neural network-based CC-RIF according to at least one embodiment of the present disclosure.

[0036] Fig.23 is a diagram showing the brightness RIF.

[0037] Fig.24 is a diagram illustrating deep learning-based CC-RIF according to at least one embodiment of the present disclosure.

[0038] Fig.25 The present invention is a flowchart of a method for upsampling a reconstructed frame performed by a video encoding device according to at least one embodiment of the present disclosure.

[0039] Fig.26 is a flowchart of a method for upsampling a reconstructed frame performed by a video decoding device according to at least one embodiment of the present invention. DETAILED DESCRIPTION

[0040] Hereinafter, some embodiments of the present disclosure are described in detail with reference to the accompanying drawings. In the following description, the same reference numerals represent the same elements, although the elements are shown in different drawings. In addition, in the following description of some embodiments, for the purpose of clarity and brevity, detailed descriptions of related known components and functions may be omitted when it is considered to obscure the subject matter of the present disclosure.

[0041] Figure 1 is a block diagram of a video encoding device that can implement the technology of the present disclosure. Figure 1 A diagram of a video encoding device and components of the device is described.

[0042] The encoding device may include a picture separator 110, a predictor 120, a subtractor 130, a transformer 140, a quantizer 145, a rearrangement unit 150, an entropy encoder 155, an inverse quantizer 160, an inverse transformer 165, an adder 170, a loop filter unit 180 and a memory 190.

[0043] Each component of the encoding device can be implemented as hardware or software or as a combination of hardware and software. In addition, the function of each component can be implemented as software, and a microprocessor can also be implemented to execute the function of the software corresponding to each component.

[0044] A video is composed of one or more sequences including multiple pictures. Each picture is divided into multiple regions, and encoding is performed on each region. For example, a picture is split into one or more tiles or / and slices. Here, one or more tiles can be defined as tile groups. Each tile or / and slice is split into one or more coding tree units (CTUs). In addition, each CTU is split into one or more coding units (CUs) through a tree structure. The information applied to each coding unit (CU) is encoded as the syntax of the CU, and the information commonly applied to the CU contained in a CTU is encoded as the syntax of the CTU. In addition, the information commonly applied to all blocks in a slice is encoded as the syntax of the slice header, and the information applied to all blocks constituting one or more pictures is encoded as a picture parameter set (PPS) or a picture header. In addition, the information commonly referenced by multiple pictures is encoded into a sequence parameter set (SPS). In addition, the information commonly referenced by one or more SPSs is encoded into a video parameter set (VPS). Further, the information commonly applied to a tile or tile group can also be encoded as the syntax of the tile or tile group header. The syntax included in the SPS, PPS, slice header, tile, or tile group header may be referred to as a high-level syntax.

[0045] The picture partitioner 110 determines the size of a coding tree unit (CTU). Information on the size of the CTU (CTU size) is encoded as a syntax of an SPS or a PPS and delivered to a video decoding device.

[0046] The picture divider 110 divides each picture constituting a video into a plurality of coding tree units (CTUs) having a predetermined size, and then recursively divides the CTUs by using a tree structure. A leaf node in the tree structure becomes a coding unit (CU), which is a basic unit of encoding.

[0047] The tree structure may be a quadtree (QT), in which a higher node (or parent node) is divided into four lower nodes (or child nodes) of the same size. The tree structure may also be a binary tree (BT), in which a higher node is divided into two lower nodes. The tree structure may also be a ternary tree (TT), in which a higher node is divided into three lower nodes at a ratio of 1:2:1. The tree structure may also be a structure in which two or more of the QT structure, the BT structure, and the TT structure are mixed. For example, a quadtree plus a binary tree (QTBT) structure may be used, or a quadtree plus a binary tree ternary tree (QTBTTT) structure may be used. Here, a binary tree ternary tree (BTTT) is added to the tree structure to be referred to as a multi-type tree (MTT).

[0048] Figure 2 is a diagram for describing a method of partitioning a block by using a QTBTTT structure.

[0049] like Figure 2 As shown, the CTU can be first split into a QT structure. The quadtree partitioning can be recursive until the size of the partitioned block reaches the minimum block size (MinQTSize) of the leaf node allowed in QT. The first flag (QT_split_flag) indicating whether each node of the QT structure is divided into four nodes of the lower layer is encoded by the entropy encoder 155 and transmitted to the video decoding device. When the leaf node of QT is not larger than the maximum block size (MaxBTSize) of the root node allowed in BT, the leaf node can be further divided into at least one of the BT structure or the TT structure. There may be multiple partitioning directions in the BT structure and / or the TT structure. For example, there may be two directions, namely, the direction in which the blocks of the corresponding node are horizontally divided and the direction in which the blocks of the corresponding node are vertically divided. As Figure 2 As shown, when MTT splitting starts, a second flag (mtt_split_flag) indicating whether a node is split and a flag further indicating the splitting direction (vertical or horizontal) and / or a flag indicating the splitting type (binary or ternary) if the node is split are encoded by the entropy encoder 155 and notified to the video decoding device with a signal.

[0050] Alternatively, before encoding the first flag (QT_split_flag) indicating whether each node is split into four nodes of the lower layer, a CU split flag (split_cu_flag) indicating whether the node is split may also be encoded. When the value of the CU split flag (split_cu_flag) indicates that each node is not split, the block of the corresponding node becomes a leaf node in the split tree structure and becomes a CU as a basic unit of encoding. When the value of the CU split flag (split_cu_flag) indicates that each node is split, the video encoding device first starts encoding the first flag through the above scheme.

[0051] When QTBT is used as another example of a tree structure, there may be two types, namely, a type in which the block of the corresponding node is horizontally divided into two blocks of the same size (i.e., symmetrical horizontal division) and a type in which the block of the corresponding node is vertically divided into two blocks of the same size (i.e., symmetrical vertical division). A split flag (split_flag) indicating whether each node of the BT structure is split into blocks of the lower layer and split type information indicating the split type are encoded by the entropy encoder 155 and delivered to the video decoding device. At the same time, a type in which the block of the corresponding node is split into two blocks that are asymmetric to each other may be presented in addition. The asymmetric form may include a form in which the block of the corresponding node is divided into two rectangular blocks having a size ratio of 1:3, or may also include a form in which the block of the corresponding node is divided in a diagonal direction.

[0052] A CU may have different sizes depending on the QTBT or QTBTTT split from a CTU. Hereinafter, a block corresponding to a CU to be encoded or decoded (i.e., a leaf node of a QTBTTT) is referred to as a "current block". Due to the QTBTTT split, the shape of the current block may be a rectangular shape in addition to a square shape.

[0053] The predictor 120 predicts the current block to generate a predicted block. The predictor 120 includes an intra predictor 122 and an inter predictor 124.

[0054] In general, each of the current blocks in a picture may be predictively encoded. Typically, the prediction of the current block may be performed using intra-prediction techniques (using data from the picture including the current block) or inter-prediction techniques (using data from a picture coded before the picture including the current block). Inter-prediction includes both unidirectional prediction and bidirectional prediction.

[0055] The intra-frame predictor 122 predicts the pixels in the current block by using the pixels (reference pixels) located on the neighbors of the current block in the current picture including the current block. According to the prediction direction, there are multiple intra-frame prediction modes. For example, as shown in FIG. 3A, multiple intra-frame prediction modes may include 2 non-directional modes including a plane mode and a DC mode, and may include 65 directional modes. Neighboring pixels and the arithmetic equations to be used are defined differently according to each prediction mode.

[0056] In order to perform effective directional prediction on the current block having a rectangular shape, a directional mode (#67 to #80, intra prediction mode #-1 to #-14) as illustrated by the dotted arrows in FIG. 3B may be additionally used. The directional mode may be referred to as a "wide-angle intra prediction mode". In FIG. 3B, the arrow indicates the corresponding reference sample used for prediction and does not indicate the prediction direction. The prediction direction is opposite to the direction indicated by the arrow. When the current block has a rectangular shape, the wide-angle intra prediction mode is a mode in which prediction is performed in a direction opposite to a specific directional mode without additional bit transmission. In this case, in the wide-angle intra prediction mode, some wide-angle intra prediction modes that can be used for the current block may be determined by the ratio of the width and height of the current block having a rectangular shape. For example, when the current block has a rectangular shape with a height less than the width, a wide-angle intra prediction mode (intra prediction mode #67 to #80) with an angle less than 45 degrees is available. When the current block has a rectangular shape with a width greater than the height, a wide-angle intra prediction mode with an angle greater than -135 degrees may be used.

[0057] The intra-frame predictor 122 may determine an intra-frame prediction to be used for encoding the current block. In some embodiments, the intra-frame predictor 122 may encode the current block by using a plurality of intra-frame prediction modes, and may further select an appropriate intra-frame prediction mode to be used from a test mode. For example, the intra-frame predictor 122 may calculate a rate-distortion value by using a rate-distortion analysis for a plurality of tested intra-frame prediction modes, and may further select an intra-frame prediction mode having the best rate-distortion characteristics in the test mode.

[0058] The intra-frame predictor 122 selects an intra-frame prediction mode from among a plurality of intra-frame prediction modes and predicts the current block by using adjacent pixels (reference pixels) and an arithmetic equation determined according to the selected intra-frame prediction mode. Information about the selected intra-frame prediction mode is encoded by the entropy encoder 155 and delivered to the video decoding device.

[0059] The inter-frame predictor 124 generates a prediction block for the current block by using a motion compensation process. The inter-frame predictor 124 searches for a block most similar to the current block in a reference picture that is encoded and decoded earlier than the current picture, and generates a prediction block for the current block by using the searched block. In addition, a motion vector (MV) is generated, which corresponds to the displacement between the current block in the current picture and the prediction block in the reference picture. Typically, motion estimation is performed on the luminance component, and the motion vector calculated based on the luminance component is used for both the luminance component and the chrominance component. Motion information including information about the reference picture and information about the motion vector for predicting the current block is encoded by the entropy encoder 155 and delivered to the video decoding device.

[0060] The inter-frame predictor 124 may also perform interpolation on a reference picture or a reference block in order to increase the accuracy of the prediction. In other words, the sub-samples between two consecutive integer samples are interpolated by applying the filter coefficients to a plurality of consecutive integer samples including two integer samples. When the process of searching for the block most similar to the current block is performed for the interpolated reference picture, the decimal unit precision may be used instead of the integer sample unit precision for the motion vector representation. The precision or resolution of the motion vector may be set differently for each target region to be encoded (e.g., units such as slices, tiles, CTUs, CUs, etc.). When this adaptive motion vector resolution (AMVR) is applied, information about the motion vector resolution to be applied to each target region should be signaled for each target region. For example, when the target region is a CU, information about the motion vector resolution applied to each CU is signaled. The information about the motion vector resolution may be information representing the precision of the motion vector difference to be described below.

[0061] Meanwhile, the inter-frame predictor 124 may perform inter-frame prediction by using bidirectional prediction. In the case of bidirectional prediction, two reference pictures and two motion vectors representing the block position most similar to the current block in each reference picture are used. The inter-frame predictor 124 selects a first reference picture and a second reference picture from a reference picture list 0 (RefPicList0) and a reference picture list 1 (RefPicList1), respectively. The inter-frame predictor 124 also searches for a block most similar to the current block in the corresponding reference picture to generate a first reference block and a second reference block. In addition, a prediction block of the current block is generated by averaging or weighted averaging the first reference block and the second reference block. In addition, motion information including information about two reference pictures for predicting the current block and including information about two motion vectors is delivered to the entropy encoder 155. Here, the reference picture list 0 may be composed of pictures before the current picture in the display order among the pre-reconstructed pictures, and the reference picture list 1 may be composed of pictures after the current picture in the display order among the pre-reconstructed pictures. However, although not particularly limited thereto, a pre-reconstructed picture after the current picture in the display order may be additionally included in the reference picture list 0. Conversely, a pre-reconstructed picture before the current picture may also be additionally included in the reference picture list 1.

[0062] In order to minimize the number of bits consumed for encoding motion information, various methods may be used.

[0063] For example, when the reference picture and motion vector of the current block are the same as those of the neighboring block, information capable of identifying the neighboring block is encoded to deliver the motion information of the current block to the video decoding device. This method is called merge mode.

[0064] In the merge mode, the inter predictor 124 selects a predetermined number of merge candidate blocks (hereinafter, referred to as “merge candidates”) from the neighboring blocks of the current block.

[0065] As a neighboring block for deriving merge candidates, such as Figure 4 As illustrated in FIG, all or some of the left block A0, the lower left block A1, the upper block B0, the upper right block B1, and the upper left block B2 adjacent to the current block in the current picture may be used. Further, blocks other than the current picture where the current block is located, which are located within a reference picture (which may be the same as or different from the reference picture used to predict the current block), may also be used as merge candidates. For example, a block located at the same location as the current block or a block adjacent to the block located at the same location in the reference picture may be additionally used as a merge candidate. If the number of merge candidates selected by the method described above is less than the preset number, a zero vector is added to the merge candidate.

[0066] The inter-frame predictor 124 configures a merge list including a predetermined number of merge candidates by using adjacent blocks. A merge candidate to be used as motion information of the current block is selected from the merge candidates included in the merge list, and merge index information for identifying the selected candidate is generated. The generated merge index information is encoded by the entropy encoder 155 and delivered to the video decoding device.

[0067] Merge skip mode is a special case of merge mode. After quantization, when all transform coefficients for entropy coding are close to zero, only the neighbor block selection information is transmitted without the residual signal. By using merge skip mode, relatively high coding efficiency can be achieved for pictures with slight motion, still pictures, screen content pictures, etc.

[0068] Hereinafter, the merge mode and the merge skip mode are collectively referred to as the merge / skip mode.

[0069] Another method for encoding motion information is the Advanced Motion Vector Prediction (AMVP) mode.

[0070] In the AMVP mode, the inter-frame predictor 124 derives a motion vector predictor candidate for a motion vector of a current block by using neighboring blocks of the current block. As neighboring blocks for deriving motion vector predictor candidates, the same as in Figure 4 All or some of the left block A0, lower left block A1, upper block B0, upper right block B1 and upper left block B2 adjacent to the current block in the current picture illustrated in the figure. In addition, blocks other than the current picture where the current block is located, which are located in a reference picture (which may be the same as or different from the reference picture used to predict the current block), may also be used as adjacent blocks for deriving motion vector prediction sub-candidates. For example, blocks co-located with the current block in the reference picture or blocks adjacent to the co-located blocks may be used. If the number of motion vector candidates selected by the above method is less than the preset number, a zero vector is added to the motion vector candidates.

[0071] The inter predictor 124 derives a motion vector predictor candidate by using the motion vectors of the neighboring blocks, and determines a motion vector predictor for the motion vector of the current block by using the motion vector predictor candidate. In addition, a motion vector difference is calculated by subtracting the motion vector predictor from the motion vector of the current block.

[0072] The motion vector predictor can be obtained by applying a predefined function (e.g., center value and average value calculation, etc.) to the motion vector predictor candidate. In this case, the video decoding device also knows the predefined function. In addition, since the neighboring blocks used to derive the motion vector predictor candidate are blocks for which encoding and decoding have been completed, the video decoding device may also already know the motion vectors of the neighboring blocks. Therefore, the video encoding device does not need to encode the information for identifying the motion vector predictor candidate. Thus, in this case, information about the motion vector difference and information about the reference picture used to predict the current block are encoded.

[0073] Meanwhile, the motion vector predictor may also be determined by selecting any one of the motion vector predictor candidates. In this case, information for identifying the selected motion vector predictor candidate is additionally encoded jointly with information about the motion vector difference and information about the reference picture used to predict the current block.

[0074] The subtractor 130 generates a residual block by subtracting a prediction block generated by the intra predictor 122 or the inter predictor 124 from the current block.

[0075] The transformer 140 converts the residual signal in the residual block having the pixel value in the spatial domain into a transform coefficient in the frequency domain. The transformer 140 may transform the residual signal in the residual block by using the total size of the residual block as a transform unit, or may further divide the residual block into a plurality of sub-blocks, and may perform the transform by using the sub-block as a transform unit. Optionally, the residual block is divided into two sub-blocks, namely a transform region and a non-transform region, so as to transform the residual signal using only the transform region sub-block as a transform unit. Here, the transform region sub-block may be one of two rectangular blocks having a size ratio of 1:1 based on the horizontal axis (or vertical axis). In this case, the flag (cu_sbt_flag) indicates that only the sub-block is transformed, and the directional (vertical / horizontal) information (cu_sbt_horizontal_flag) and / or the position information (cu_sbt_pos_flag) are encoded by the entropy encoder 155 and signaled to the video decoding device. In addition, the size of the transform region sub-block may have a size ratio of 1:3 based on the horizontal axis (or vertical axis). In this case, a flag (cu_sbt_quad_flag) of partitioning the corresponding partition is additionally encoded by the entropy encoder 155 and notified to the video decoding apparatus by a signal.

[0076] Meanwhile, the transformer 140 may perform transformation on the residual block separately in the horizontal direction and the vertical direction. For the transformation, different types of transformation functions or transformation matrices may be used. For example, a pair of transformation functions for horizontal transformation and vertical transformation may be defined as a multiple transformation set (MTS). The transformer 140 may select a transformation function pair with the highest transformation efficiency in the MTS, and may transform the residual block in each of the horizontal and vertical directions. Information (mts_idx) of the transformation function pair in the MTS is encoded by the entropy encoder 155 and notified to the video decoding device with a signal.

[0077] The quantizer 145 quantizes the transform coefficients output from the transformer 140 using the quantization parameter and outputs the quantized transform coefficients to the entropy encoder 155. The quantizer 145 can also immediately quantize the relevant residual block without transforming any block or frame. The quantizer 145 can also apply different quantization coefficients (scaling values) according to the position of the transform coefficient in the transform block. The quantization matrix applied to the quantized transform coefficients arranged in 2 dimensions can be encoded and notified to the video decoding device with a signal.

[0078] The rearrangement unit 150 may perform rearrangement of coefficient values ​​for the quantized residual value.

[0079] The rearrangement unit 150 may change the 2D coefficient array into a 1D coefficient sequence by using coefficient scanning. For example, the rearrangement unit 150 may output an ID coefficient sequence by scanning the DC coefficient into a high frequency domain coefficient using a zigzag scan or a diagonal scan. Depending on the size of the transform unit and the intra prediction mode, a vertical scan that scans the 2D coefficient array in the column direction and a horizontal scan that scans the 2D block type coefficients in the row direction may also be used instead of the zigzag scan. In other words, depending on the size of the transform unit and the intra prediction mode, the scanning method to be used may be determined among zigzag scanning, diagonal scanning, vertical scanning, and horizontal scanning.

[0080] The entropy encoder 155 generates a bitstream by encoding the sequence of ID-quantized transform coefficients output from the rearrangement unit 150 using various encoding schemes including context-based adaptive binary arithmetic coding (CABAC), exponential Golomb, and the like.

[0081] In addition, the entropy encoder 155 encodes information related to block partitioning (such as CTU size, CTU split flag, QT split flag, MTT partition type, MTT partition direction, etc.) to allow the video decoding device to equally partition the block to the video encoding device. In addition, the entropy encoder 155 encodes information about the prediction type indicating whether the current block is encoded by intra-frame prediction or inter-frame prediction. The entropy encoder 155 encodes intra-frame prediction information (i.e., information about intra-frame prediction mode) or inter-frame prediction information (in the case of merge mode, merge index, and in the case of AMVP mode, information about reference picture index and motion vector difference) according to the prediction type. In addition, the entropy encoder 155 encodes information related to quantization (i.e., information about quantization parameters and information about quantization matrices).

[0082] The inverse quantizer 160 dequantizes the quantized transform coefficient output from the quantizer 145 to generate a transform coefficient. The inverse transformer 165 transforms the transform coefficient output from the inverse quantizer 160 from the frequency domain to the spatial domain to reconstruct a residual block.

[0083] The adder 170 reconstructs the current block by adding the reconstructed residual block and the prediction block generated by the predictor 120. When intra prediction is performed on a next sequential block, pixels in the reconstructed current block may be used as reference pixels.

[0084] The loop filter unit 180 performs filtering on the reconstructed pixels in order to reduce block artifacts, ringing artifacts, blur artifacts, etc. that occur due to block-based prediction and transform / quantization. The loop filter unit 180 as a loop filter may include all or some of the deblocking filter 182, the sample adaptive offset (SAO) filter 184, and the adaptive loop filter (ALF) 186.

[0085] The deblocking filter 182 filters the boundaries between the reconstructed blocks to remove the block effect caused by block unit encoding / decoding, and the SAO filter 184 and the ALF 186 perform additional filtering for the deblocked filtered video. The SAO filter 184 and the ALF 186 are filters for compensating the difference between the reconstructed pixels and the original pixels, which is caused by lossy coding. The SAO filter 184 applies an offset as a CTU unit to enhance subjective picture quality and coding efficiency. On the other hand, the ALF 186 performs block unit filtering and applies different filters to compensate for distortion by dividing the boundaries of the corresponding blocks and the degree of the change amount. Information about the filter coefficients to be used for the ALF can be encoded and notified to the video decoding device with a signal.

[0086] The reconstructed blocks filtered by the deblocking filter 182, the SAO filter 184, and the ALF 186 are stored in the memory 190. When all blocks in one picture are reconstructed, the reconstructed picture can be used as a reference picture for inter-frame prediction of blocks in a picture to be encoded later.

[0087] Figure 5 is a functional block diagram of a video decoding device that can implement the techniques of the present invention. Figure 5 , describing a video decoding apparatus and components of the apparatus.

[0088] The video decoding device may include an entropy decoder 510 , a rearrangement unit 515 , an inverse quantizer 520 , an inverse transformer 530 , a predictor 540 , an adder 550 , a loop filter unit 560 , and a memory 570 .

[0089] and Figure 1 Similar to the video encoding device of the present invention, each component of the video decoding device can be implemented as hardware or software or as a combination of hardware and software. In addition, the function of each component can be implemented as software, and a microprocessor can also be implemented to execute the function of the software corresponding to each component.

[0090] The entropy decoder 510 extracts information related to block partitioning by decoding a bitstream generated by a video encoding apparatus to determine a current block to be decoded, and extracts prediction information required to reconstruct the current block and information about a residual signal.

[0091] The entropy decoder 510 determines the size of the CTU by extracting information about the CTU size from a sequence parameter set (SPS) or a picture parameter set (PPS), and divides the picture into CTUs of a determined size. In addition, the CTU is determined to be the highest level of the tree structure, i.e., the root node, and the partition information of the CTU can be extracted to partition the CTU using the tree structure.

[0092] For example, when the CTU is divided using the QTBTTT structure, the first flag (QT_split_flag) related to the division of the QT is first extracted to divide each node into four nodes of the lower layer. In addition, for the node corresponding to the leaf node of the QT, the second flag (mtt_split_flag) related to the division of the MTT, the division direction (vertical / horizontal) and / or the division type (binary / ternary) are extracted to divide the corresponding leaf node into the MTT structure. As a result, each node below the leaf node of the QT is recursively split into a BT or TT structure.

[0093] As another embodiment, when a CTU is divided by using a QTBTTT structure, a CU split flag (split_cu_flag) indicating whether the CU is divided is extracted. When the corresponding block is divided, a first flag (QT_split_flag) may also be extracted. During the splitting process, for each node, 0 or more recursive MTT splits may occur after 0 or more recursive QT splits. For example, for a CTU, MTT splitting may occur immediately, or conversely, only multiple QT splits may occur.

[0094] For another example, when the CTU is split using the QTBT structure, the first flag (QT_split_flag) related to the splitting of the QT is extracted to split each node into four nodes of the lower layer. In addition, the split flag (split_flag) indicating whether the node corresponding to the leaf node of the QT is further split into BTs and the split direction information are extracted.

[0095] Meanwhile, when the entropy decoder 510 determines the current block to be decoded by using the partitioning of the tree structure, the entropy decoder 510 extracts information about the prediction type indicating whether the current block is intra-predicted or inter-predicted. When the prediction type information indicates intra-prediction, the entropy decoder 510 extracts syntax elements of intra-prediction information (intra-prediction mode) for the current block. When the prediction type information indicates inter-prediction, the entropy decoder 510 extracts information representing syntax elements for inter-prediction information (i.e., motion vectors and reference pictures referenced by motion vectors).

[0096] Furthermore, the entropy decoder 510 extracts quantization-related information and extracts information on a quantized transform coefficient of the current block as information on a residual signal.

[0097] The rearrangement unit 515 may change the sequence of 1D quantized transform coefficients entropy-decoded by the entropy decoder 510 into a 2D coefficient array (ie, block) again in the reverse order of the coefficient scanning order performed by the video encoding device.

[0098] The inverse quantizer 520 dequantizes the quantized transform coefficients and dequantizes the quantized transform coefficients by using quantization parameters. The inverse quantizer 520 may also apply different quantization coefficients (scaling values) to the quantized transform coefficients arranged in 2D. The inverse quantizer 520 may perform inverse quantization by applying a matrix (scaling value) of quantization coefficients from a video encoding device to a 2D array of quantized transform coefficients.

[0099] The inverse transformer 530 generates a residual block for the current block by reconstructing a residual signal by inversely transforming the dequantized transform coefficients from the frequency domain into the spatial domain.

[0100] In addition, when the inverse transformer 530 inversely transforms a partial region (sub-block) of the transform block, the inverse transformer 530 extracts a flag (cu_sbt_flag) that only the sub-block of the transform block is transformed, direction (vertical / horizontal) information (cu_sbt_horizontal_flag) of the sub-block, and / or position information (cu_sbt_pos_flag) of the sub-block. The inverse transformer 530 also inversely transforms the transform coefficient of the corresponding sub-block from the frequency domain to the spatial domain to reconstruct the residual signal and fills the region that is not inversely transformed with a value of "0" as the residual signal to generate a final residual block for the current block.

[0101] In addition, when MTS is applied, the inverse transformer 530 determines a transform index or a transform matrix applied in each of the horizontal and vertical directions by using MTS information (mts_idx) signaled from the video encoding device. The inverse transformer 530 also performs inverse transform on the transform coefficients in the transform block in the horizontal and vertical directions by using the determined transform function.

[0102] The predictor 540 may include an intra predictor 542 and an inter predictor 544. The intra predictor 542 is activated when the prediction type of the current block is intra prediction, and the inter predictor 544 is activated when the prediction type of the current block is inter prediction.

[0103] The intra predictor 542 determines an intra prediction mode of a current block among a plurality of intra prediction modes according to a syntax element of the intra prediction mode extracted from the entropy decoder 510. The intra predictor 542 also predicts the current block by using neighboring reference pixels of the current block according to the intra prediction mode.

[0104] The inter predictor 544 determines a motion vector of a current block and a reference picture to which the motion vector refers by using a syntax element for the inter prediction mode extracted from the entropy decoder 510 .

[0105] The adder 550 reconstructs the current block by adding the residual block output from the inverse transformer 530 and the prediction block output from the inter predictor 544 or the intra predictor 542. When intra prediction is performed on a block to be decoded later, pixels within the reconstructed current block are used as reference pixels.

[0106] The loop filter unit 560 as a loop filter may include a deblocking filter 562, an SAO filter 564, and an ALF 566. The deblocking filter 562 performs deblocking filtering on the boundary between the reconstructed blocks in order to remove the block effect that occurs due to block unit decoding. The SAO filter 564 and the ALF 566 perform additional filtering on the reconstructed blocks after deblocking filtering in order to compensate for the difference between the reconstructed pixels and the original pixels that occurs due to lossy encoding. The filter coefficients of the ALF are determined by using information about the filter coefficients decoded from the bitstream.

[0107] The reconstructed blocks filtered by the deblocking filter 562, the SAO filter 564, and the ALF 566 are stored in the memory 570. When all blocks in one picture are reconstructed, the reconstructed picture can be used as a reference picture for inter-frame prediction of blocks in a picture to be encoded later.

[0108] In some embodiments, the present invention relates to encoding and decoding video images as described above. More specifically, the present invention provides a video encoding method and apparatus that utilizes an in-loop filter based on upsampling or super-resolution technology to reconstruct chrominance components.

[0109] The following embodiments may be performed by the loop filter unit 180 in the video encoding apparatus. The following embodiments may also be performed by the loop filter unit 560 in the video decoding apparatus.

[0110] The video encoding device in the in-loop filtering of the chroma component can generate signaling information associated with the present embodiment in terms of rate-distortion optimization. The video encoding device can encode the signaling information using the entropy encoder 155 and transmit the encoded signaling information to the video decoding device. The video decoding device can decode the signaling information associated with the in-loop filtering of the chroma component from the bitstream using the entropy decoder 510.

[0111] In the following description, the term 'target block' may be used interchangeably with a current block or a coding unit (CU), or may refer to some regions of a coding unit.

[0112] In addition, a flag value of true indicates when the flag is set to 1. In addition, a flag value of false indicates when the flag is set to 0.

[0113] I. Convolutional Neural Network (CNN)

[0114] CNN refers to a neural network composed of multiple convolutional and pooling layers and is known to be the best deep learning technique for image processing. The convolutional layer uses multiple kernels or filters to extract feature maps (also called "features"). The kernel coefficients that make up the filter are parameters determined during the learning process.

[0115] In the convolutional layers of a CNN, the front layers close to the input extract feature maps responsive to simple low-level image features (such as lines, points, or surfaces), while the back layers close to the output extract feature maps responsive to high-level features (such as textures and object parts).

[0116] Figure 6 is a diagram illustrating the operation of a convolutional layer according to at least one embodiment of the present disclosure.

[0117] The convolutional layer uses convolution operations to generate feature maps from the input image. Figure 6 A kernel (or filter) having a kernel size of 3 x 3 is shown. The kernel size is also referred to as the filter size. The kernel has kernel parameters or filter parameters, also referred to as weights. Figure 6 The kernel shown in has a total of nine kernel parameters. The kernel parameters can be initially set to arbitrary values, and the values ​​of the parameters can be updated based on learning.

[0118] The convolution layer performs a convolution operation by using a block of size equal to the kernel size in the input image. In this case, the block of size of the kernel in the input image is called a window.

[0119] When filtering the input image in raster scan order, the displacement of the window is called the stride. Figure 6 In the embodiment of , the stride is 1. If the stride is set to 2, the convolution operation is performed with a window offset by 2 samples, resulting in the horizontal and vertical dimensions of the feature map being half of the horizontal and vertical dimensions of the input image.

[0120] As mentioned above, a convolutional layer may include multiple filters. The number of filters or the number of kernels is called channels. Therefore, the number of channels is equal to the number of filters. The number of filters also determines the dimensionality of the feature map.

[0121] Padding refers to a method of expanding the input data by filling specific values ​​around the input data before performing the convolution operation. Padding is mainly used to adjust the spatial size of the output data. The value used for padding can be determined by a hyperparameter, but zero padding is usually used. If padding is not used, the spatial size of the output data decreases with each convolution layer, which may cause boundary information to be lost. Therefore, as a remedy, padding is used. Therefore, padding can be used to equalize the spatial size between the output data of the convolution layer and the input data.

[0122] Figure 7 is a diagram illustrating the operation of a deconvolution layer according to at least one embodiment of the present disclosure.

[0123] The deconvolution layer performs the opposite operation of the convolution layer. The deconvolution layer takes the feature map as input and generates the desired data image as output. The deconvolution layer uses the same terminology as the convolution layer. Figure 7As shown in the embodiment of , when the stride is 1, the horizontal dimension and vertical dimension of the feature map are the same as the horizontal dimension and vertical dimension of the output data. When the stride is 2, the width and height of the output data are twice the width and height of the feature map.

[0124] Figure 8 is a diagram illustrating the operation of a pooling layer according to at least one embodiment of the present disclosure.

[0125] The pooling layer performs pooling, which is the process of subsampling the feature maps generated by the convolutional layer. The pooling layer uses a 2×2 window to select samples so that the output is half the horizontal and vertical size of the input, respectively. In other words, the pooling layer is used to reduce the size of the input image or input feature map by aggregating 2×2 regions into a single sample. Figure 8 As shown, the pooling methods include maximum pooling that selects the maximum value in a 2×2 area and average pooling that generates the average value of a 2×2 area. Unlike the convolutional layer, the pooling layer does not include variables that need to be trained and keeps the number of channels in the input the same as the number of channels in the output.

[0126] The opposite of the pooling layer is defined as the up-pooling layer. The up-pooling layer acts as the opposite of the pooling layer, increasing the dimension, and is usually used after the deconvolution layer.

[0127] The convolutional encoder-decoder structure is a network structure consisting of pairs of convolutional layers and deconvolutional layers. The convolutional encoder consists of convolutional layers and pooling layers to output feature maps (or feature vectors) from the input image. The final output vector of the convolutional encoder is also called a latent vector. The convolutional decoder consists of deconvolutional layers and unpacking layers to generate an output image from the feature map or latent vector.

[0128] The input and output of the convolutional encoder-decoder can be variably set according to the application and the purpose of the network. For example, the input and output can be an optical flow map, a saliency map, an image frame, etc.

[0129] Fig. 9 is a diagram showing a SISR network.

[0130] An example of an application of CNN is single image super-resolution (SISR). A SISR network generates a high-resolution image as output from a low-resolution input image. Fig. 9 As shown, the SISR network may include multiple convolutional layers. Each convolutional layer includes an activation function, such as a rectified linear unit (ReLU). The parameters of the SISR network can be trained so that the resulting super-resolution (SR) image is close to the ground truth (GT).

[0131] SR methods using CNN can improve SR performance by increasing the depth (meaning, for example, by increasing the number of convolutional layers). To overcome the problem of overfitting that may occur in learning as the depth increases, residual blocks that can perform skip connections and residual learning can be used in SISR networks. Fig.10 As shown, the residual block is used to apply the convolution operation to the input feature x l In addition, the residual block can generate the output x l+1 The path to apply convolution operation or the skip path is selected based on learning efficiency. Fig.10 In an embodiment of the present invention, the residual block includes a batch normalization (BN) layer.

[0132] In one embodiment, an enhanced deep residual network (EDSR) for SISR increases the performance of the network by increasing the depth by serially concatenating residual blocks. Another example of accurate image super-resolution using a very deep convolutional network (VDSR) is a visual geometry group or VGG network-based CNN model that uses residual learning and adds a residual frame to the final output based on residual learning of the residual signal. VDSR adds the residual signal to the input signal by adding the residual signal at the very end of the network.

[0133] II. Adaptive Loop Filter (ALF)

[0134] The ALF 186, 566 of VVC uses an adaptive linear filter based on the Wiener-Hopf equation to approximate the reconstructed video frame to the original frame. The video encoding device uses the output samples of the SAO filter 184 to calculate the filter coefficients of the ALF 186 based on rate-distortion optimization and transmits them to the video decoding device. Fig.11 As shown in the embodiment of , ALF 186, 566 is configured as a 7×7 diamond shape and a 5×5 diamond shape for luminance and chrominance sampling, respectively. The filter shape and size can be determined by considering the trade-off between coding efficiency and computational complexity. For example, using a symmetric FIR filter can reduce the computational complexity of ALF 186, 566.

[0135] To export in Fig.11 The filter coefficients c shown in i , using the samples at the corresponding positions. The filtered samples I(x, y) at the current position (x, y) can be calculated by 7-bit precision operations as shown in Equation 1.

[0136] [Equation 1]

[0137]

[0138] Here, r iis the difference between the current sample and the adjacent sample calculated according to Equation 2.

[0139] [Equation 2]

[0140] r i =min(b i ,max(-b i , I(x+x i ,y+y i )-I(x,y)))

[0141] +min(b i ,max(-b i ,I(xx i ,yy i )-I(x,y)))

[0142] Here, b i is the clipping parameter.

[0143] ALF 186, 566 utilizes up to 25 sets of filter coefficients for the luminance component, and applies the 25 sets of filter coefficients to the 4×4 sub-block. The 4×4 sub-block is classified into one of 25 categories based on the gradient information of the local block calculated by using the Laplacian filter. Specifically, the category classification index is derived from a combination of five directional attributes representing the intensity and direction of the texture component and five activity attributes of the sub-block. In addition, geometric transformations such as 90-degree rotation, diagonal shift, and vertical shift may be applied to the filter coefficients before filtering. By using geometric transformations to take into account different directionality, a wider variety of block characteristics can be handled by using a smaller set of filter coefficients.

[0144] In addition to the sub-block level, the decision to apply can be made at the CTU level. For chroma components, up to eight filters are used at the CTU level. The chroma ALF is activated only if the luma ALF is activated at that level.

[0145] Meanwhile, an adaptive parameter set (APS) is used to convey ALF filter parameters, including a set of filter coefficients. As described above, up to 25 sets of filter coefficients can be calculated for luma components, and up to 8 sets of filter coefficients can be calculated for chroma components. If the same ALF coefficients are used for different slices, the index of the reference APS can be signaled instead of redundantly retransmitting the ALF coefficients.

[0146] In video applications such as High Dynamic Range (HDR) and Wide Color Gamut (WCG), reconstruction of video colors is critical. Cross-Component ALF (CC-ALF) uses the correlation between the current chroma sample and luma sample at that position to modify the chroma samples in parallel with ALF.

[0147] Fig.12 is a diagram showing application of a cross-component adaptive loop filter (CC-ALF).

[0148] To generate corrected values ​​of chroma samples from the input luma samples, such as Fig.12 As shown in the example of , CC-ALF performs a linear filtering operation. The linear filtering operation uses the luminance samples (R Y ()) as input and generates a correlation value (ΔR i ()), as shown in Eq. 3.

[0149] [Equation 3]

[0150]

[0151] Here, (x, y) is the position of each chroma sample, and (x c ,y c ) is the location of the brightness sample corresponding to (x, y). 0 ,y 0 ) means around (x c ,y c ) filters support offset, and c i (x 0 ,y 0 ) represents the filter coefficient. S i represents the filter target area in luma. This correlation value is then used as a correction value for improving the output of the chroma ALF, as Fig.12 shown.

[0152] In one example, a 3×4 diamond high pass filter (such as Fig.13 In the example of FIG. 1 , a luminance sample is applied to generate a correction value. In order to generate a correction value for each chroma sample, a luminance sample that has been subjected to an SAO filter (sample adaptive offset filter) corresponding to each chroma sample position is used.

[0153] The video encoding device may determine four sets of filter coefficients for CC-ALF for each chrominance component. Unlike conventional ALF, CC-ALFs filter coefficients are not symmetric constrained, but have the following characteristics. First, the sum of CC-ALF coefficients is zero. Second, the absolute value of CC-ALF coefficients is 0 or a power of 2.

[0154] The video encoding device signals one of the four sets per chroma component on a CTU basis. The CC-ALF filter coefficients may be transmitted together with the ALF parameters (Adaptive Parameter Set) of the APS. For CC-ALF used at the sequence level, ALF also needs to be used in the sequence. Similarly, for CC-ALF used at the slice or picture level, ALF also needs to be used in the slice or picture.

[0155] On the other hand, in order to save line buffers at a given filter size for luma and chroma ALF, the virtual boundaries for luma and chroma are four lines and two lines above the CTU boundary, respectively. When CC-ALF is applied to the 4:2:0 chroma format, there is no problem with the alignment of the luma and chroma line buffers based on the position of the virtual boundaries. However, when CC-ALF is applied to the 4:2:2 or 4:4:4 chroma format, due to the difference in the position of the virtual boundaries, the luma line buffer and the chroma line buffer are not aligned with each other for the 3rd and 4th lines above the CTU boundary. Therefore, in the case of 4:2:2 and 4:4:4 chroma formats, CC-ALF is not applied to samples in the 3rd and 4th lines above the CTU boundary.

[0156] III. Reference Picture Resampling (RPR)

[0157] In real-time video communication, Adaptive Resolution Change (ARC) is required due to adaptive video resolution and fast stream start, etc. Thus, the development of VVC Reference Picture Resampling (RPR) technology may refer to a picture with a different resolution than the current picture.

[0158] Fig.14 It is a block diagram of a video decoding device using the RPR technology.

[0159] The video frames downsampled by the video encoding device are encoded and transmitted, and the video frames previously decoded by the video decoding device are stored in the decoded picture buffer (DPB) in the memory 570. If the currently reconstructed video frame has a size different from the reference picture, the resampler 1410 in the video decoding device resamples (enlarges or reduces) the reference picture according to the ratio, and in the video encoding device, the motion vector is also scaled according to the ratio in the motion prediction and compensation process. The decoded video frame is stored in the DPB without resampling. When the video frame is output, the upsampler 1420 in the video decoding device upsamples the downsampled video frame to reconstruct the video frame to its original size. However, upsampling may cause the loss of high-frequency components, and therefore the output video frame may suffer from blur.

[0160] Hereinafter, embodiments of the present disclosure are described with reference to a video decoding apparatus, but they are similarly applicable to a video encoding apparatus.

[0161] IV. Implementations According to the Present Disclosure

[0162] <Implementation Method 1> Chroma Component In-Loop Filter

[0163] In this embodiment, a chroma component in-loop filter (CC-IF) enhances the video quality of the reconstructed frames based on deep learning techniques.

[0164] In the video decoding device according to this embodiment, the chroma component resampling in-loop filter (C-RIF) 1910 can replace any filter among the deblocking filter 562, sampling adaptive offset (SAO) filter 564 and adaptive loop filter (ALF) 566 that constitute the conventional chroma component in-loop filter, or it can be inserted in the middle of the conventional chroma component in-loop filter.

[0165] In addition to the low resolution (LR) chroma components, conventional chroma intra-loop filters also utilize pre-reconstructed low resolution luma components. In order to utilize luma sampling, a cascade method of luma and chroma components is used. In this embodiment, the following method is used to utilize luma sampling.

[0166] Fig.15 is a diagram illustrating pixel shuffling according to at least one embodiment of the present disclosure.

[0167] One example may utilize pixel shuffling between luma and chroma components. To this end, CC-IF may include a pixel shuffling layer. Fig.15 As shown in the example, when the dimensions are H×W×C×t 2 When a feature map of is given as input, the pixel shuffling layer generates a high-resolution image of size tH × tW × C. Here, t corresponds to the scaling value between the feature maps before and after applying CC-IF.

[0168] When the luma component is used, a feature map generated by convolution of the luma component and a feature map generated by convolution of the chroma component are applied to the pixel shuffling layer. Fig.16 As shown in the example of , the pixel shuffling layer rearranges the pixels of the feature maps of the luminance component and the chrominance component to generate a high-resolution image enlarged by a preset t. Then, the enlarged high-resolution image can be used as input to operate a conventional chrominance loop filter. Fig.16 In the example of , Ci represents the chrominance component and Ci∈{Cb,Cr}.

[0169] Fig.17 is a diagram illustrating channel attention according to at least one embodiment of the present disclosure.

[0170] As another example, a conventional chroma intra-loop filter may operate after adding channel attention to the luminance and chroma components cascaded as described above. Channel attention maximizes the interdependencies between channels in a feature map of size H×W×C. To this end, CC-IF may include a channel attention layer.

[0171] like Fig.17 As shown in the example of GP , convolutional layers, sigmoid functions, and all or part of the element-wise multiplication. Fig.17 In the example, block f represents the sigmoid function.

[0172] The channel attention layer converts H GP Applied to a feature map of size H × W to generate an initial channel attention of size 1 × 1 × C. The channel attention layer will then have learnable parameters W D and W U The convolutional layer of and the sigmoid function are applied to the initial channel attention, resulting in a final channel attention of size 1×1×C. In addition, the channel attention layer can generate a feature map of size H×W×C reflecting the attention by performing an element-wise product between the original input and the final channel attention using a skip connection. This means that the channel attention layer can reweight the input feature map by using channel attention.

[0173] As described above, CC-IF uses a channel attention layer to generate channel attention between the luminance component feature map and the chrominance component feature map and reweights each feature map according to the channel attention. The reweighted feature map can then be applied to the chrominance intra-loop filter.

[0174] Fig.18 is a diagram illustrating a chroma component in-loop filter according to at least one embodiment of the present disclosure.

[0175] After constructing the input feature map as described above, the chroma component in-loop filter generates an improved chroma component by using a reconstruction model for image quality reconstruction and enhancement. Fig.18 In the example of , the input feature map generated by the pixel shuffle layer is input to the reconstruction model. Here, the reconstruction model can be a Fig.10 A deep neural network with a cascade structure of residual blocks shown in .

[0176] The operation of CC-IF is controlled by an on / off flag at a block level (e.g., CU or CTU, slice / picture level, and video sequence level). In this case, CC-IF can be controlled according to the operation of the in-loop filter that improves the quality of the luma signal. That is, it is determined whether CC-IF is applied or not only when the luma in-loop filter is applied. Alternatively, CC-IF can be controlled separately from the luma in-loop filter. That is, the flag that controls the luma in-loop filter can be applied separately from the flag that controls CC-IF.

[0177] <Implementation Method 2> Chroma Component Resampling In-Loop Filter

[0178] The 4:4:4 video format has a chroma component that is equal in block size to a luminance component. However, encoding techniques in existing video encoding devices are mostly optimized for the 4:2:0 video format. The implementation involves techniques for downsampling chroma components, converting chroma components to a 4:2:0 format, encoding chroma components, and reconstructing chroma components back to a 4:4:4 format when a video in a 4:4:4 format is input.

[0179] In this embodiment, the chroma component resampling in-loop filter (C-RIF) performs upsampling (or super-resolution processing) of the chroma components to reconstruct the chroma components in a 4:4:4 format. The implementation is described below in a 4:4:4 format, but the implementation can also be applied to videos in a 4:2:0 format. That is, the downsampled chroma blocks are downsampled again, and then the twice downsampled chroma blocks are upsampled back to reconstruct the chroma blocks according to the previously downsampled resolution.

[0180] Fig.19 is a schematic diagram showing a video decoding apparatus including a chroma component resampling in-loop filter according to at least one embodiment of the present disclosure.

[0181] In the video decoding apparatus according to at least one embodiment, the C-RIF 1910 may replace any one of the deblocking filter 562, the SAO filter 564, and the adaptive loop filter (ALF) 566 constituting the conventional chroma component in-loop filter, or may be inserted in the middle of the conventional chroma component in-loop filter. Fig.19 In the example of , C-RIF 1910 is located between SAO filter 564 and ALF 566 .

[0182] When downsampled and compressed chroma components are input, C-RIF 1910 upsamples the compressed input and reconstructs it to its original resolution. Here, the original resolution refers to the resolution of the chroma components in the original frame. If the input video comes in at the original resolution or if the flag to perform upsampling is not enabled, no upsampling is performed by C-RIF 1910.

[0183] In this case, a technique applicable to the VVC RPR technique may be used as a filter to downsample the original frame of the original resolution. Specifically, the downsampling may be performed by using a low-pass filter using a windowed synchronization function. Further, during downsampling, the horizontal and vertical dimensions of the original frame are divided by the scaling value. In this case, the fractional unit is truncated or rounded. For example, if the size in the horizontal direction is W and the scaling value is S, the length of the downsampling is represented as W. d =floor(W / S). Here, floor is a floor function.

[0184] Rather than simply dividing by the scaling value, pixels may be padded at frame boundaries so that the downsampled result is an even multiple of four.

[0185] C-RIF 1910 upsamples the chroma component currently being reconstructed. In this case, the upsampling scaling is a factor of 2 horizontally or vertically. The scaled values ​​of the original chroma components are signaled to reconstruct the chroma components at the original resolution according to the upsampling. C-RIF 1910 can upsample the video frame currently being reconstructed by using any of the following techniques.

[0186] In one example, C-RIF 1910 uses a finite impulse response (FIR) filter to upsample the video frame. Specifically, an interpolation filter applied to generate a prediction block during inter-frame prediction can be used for upsampling.

[0187] As another example, C-RIF 1910 uses a neural network super-resolution (NNSR) technique to upsample video frames, which performs upsampling based on a neural network. Specifically, a transposed convolution may be used, or a convolution process may be performed after upsampling. By using a transposed convolution, i.e., the inverse of the mathematical operation used in convolution, the image is upsampled, and the kernel used may be updated by using learning. It may be based on Figure 7 The deconvolution layer shown in implements transposed convolution. In addition, the network structure and component techniques for super-resolution processing (such as SISR and EDSR mentioned above) can be used as NNSR.

[0188] First, the transposed convolution operation can be implemented as Fig. 20 . Fig. 20 In the transposed convolution operation shown in , C-RIF 1910 can utilize a 3×3 kernel, stride 2, and padding 1 to generate an output with increased horizontal / vertical dimensions from the input feature map. Fig. 20 As shown in the example of , a single pixel of the input feature map can be multiplied by a 3×3 kernel to generate a corresponding set of output pixels.

[0189] In addition, C-RIF 1910 can perform upsampling by using the non-pooling method described above. Figure 8 The unpooling methods that are the inverse of the pooling method shown in include maximum unpooling and average unpooling. Maximum unpooling remembers the position index of the maximum value in the matrix during maximum pooling, and then places the sampled value at the position indicated by the position index during unpooling. In this case, maximum unpooling can fill adjacent pixels with the same value, or can fill adjacent pixels with zeros. At the same time, average unpooling performs upsampling under the assumption that the input value is the average of the output value.

[0190] like Fig.19 As shown in the example of , the location of the C-RIF 1910 may be anywhere within the loop filter unit 560 in the video decoding apparatus. That is, the C-RIF 1910 may be located at the deblocking filter 562, the SAO filter 564, or the ALF 566.

[0191] For example, the C-RIF 1910 may be located after the SAO filter 564 and before the ALF 566, as shown in FIG. Fig.19 For example, the video decoding apparatus may input a signal generated by applying an FIR-based up-sampling filter to the down-sampled output samples that have passed through the SAO filter 564 to the ALF 566. Here, the input signal of the ALF 566 is represented by I(x, y).

[0192] The filter coefficients of the ALF 566 may be calculated by using the Winer-hopf equation. Specifically, after the input signal I(x, y) passes through the ALF 566, the filter coefficients for the input signal I(x, y) may be derived to approximate the original image. The conventional ALF 566 removes quantization noise between input video frames and output video frames having the same resolution, but in this embodiment, the input chrominance components and the output chrominance components have different resolutions, so in addition to the quantization noise, the filter coefficients may be derived to mitigate the blurring effect caused by the upsampling process. Although the same structure and the same filter coefficient derivation method are used as the conventional ALF 566, the addition of the C-RIF 1910 as described above may replace the use of the ALF 566.

[0193] In this case, ALF 566 may be a two-dimensional linear filter organized in a diamond or square shape. Fig.11 As shown in the example of , the ALF 566 can be configured as a 7×7 diamond shape and a 5×5 diamond shape for luma sampling and chroma sampling, respectively. The filter shape and size can be determined by considering the tradeoff between coding efficiency and computational complexity. For example, the computational complexity of the ALF 566 can be reduced by using a symmetric FIR filter.

[0194] On the other hand, whether to apply C-RIF 1910 may be determined at the level of a sequence, a picture, a sub-picture, a slice, a tile, and / or a CTU.

[0195] In one example, C-RIF 1910 depends on ALF 566 used at the relevant applicable unit level. For example, if ALF 566 is used at the relevant applicable unit, then C-RIF 1910 may also be used. For example, for C-RIF 1910 used at the slice level, ALF 566 also needs to be used in the slice.

[0196] Alternatively, the C-RIF 1910 may be used independently of the ALF 566. The flag indicating the use of the C-RIF 1910 may be decoded by the video decoding device regardless of the value of the flag indicating the use of the ALF 566.

[0197] <Implementation 3> Cross-component resampling in-loop filter

[0198] In this implementation, a cross-component resampling in-loop filter (CC-RIF) uses luma samples to enhance the upsampled chroma samples when upsampling the downsampled input chroma components and reconstructing the input chroma components to their original resolution. In addition, the same upsampling scaling as the luma component is applied to the chroma components. CC-RIF is not performed if the input video comes in at its original resolution or the flag to perform CC-RIF is not enabled.

[0199] CC-RIF can replace any of the deblocking filter 562, SAO filter 564, and ALF 566 constituting the conventional in-loop filter of the chroma component, or can be inserted in the middle of the conventional in-loop filter of the chroma component. CC-RIF can upsample the chroma sampling of the current reconstructed video frame by using any of the following techniques.

[0200] Fig.21 is a diagram illustrating application of a cross-component resampling in-loop filter in accordance with at least one embodiment of the present disclosure.

[0201] In one example, CC-RIF 2110 may improve chroma sampling by using linear filtering. That is, CC-RIF 2110 processes the luma component ( Fig.21 In order to generate correction values ​​for chroma samples upsampled from the input luma samples, CC-RIF 2110 performs a linear filtering operation, such as in Fig.21 The linear filtering operation uses the brightness samples (R Y()) as input to generate the correlation value (ΔR) with each chroma sample (i∈{Cb,Cr}) i ()), as in Eq. 4.

[0202] [Equation 4]

[0203]

[0204] Here, (x, y) is the position of each chroma sample, and (x c ,y c ) is the location of the brightness sample corresponding to (x, y). 0 ,y 0 ) means around (x c ,y c ) filters support offset, and c i (x 0 ,y 0 ) denotes the filter coefficients. Si denotes the filter target region in luminance. The correlation values ​​may be upsampled and then used as correction values ​​to improve the output of C-RIF 1910, as in Fig.21 as shown in the example.

[0205] As mentioned above, in order to save the line buffer by considering the filter size of the luma and chroma ALFs, the virtual boundaries of luma and chroma are four lines and two lines above the CTU boundary, respectively. Therefore, when CC-RIF2110 is used based on CTU, the line buffers may not be aligned for each color format at these virtual boundaries. If CC-RIF 2110 is applied to the 4:2:0 chroma format, there will be no problem with the alignment of the luma and chroma line buffers based on the position of the virtual boundary. However, when CC-RIF 2110 is applied to the 4:2:2 or 4:4:4 chroma format, due to the difference in the position of the virtual boundary, the luma line buffer and the chroma line buffer are not aligned with each other for the 3rd and 4th lines above the CTU boundary. Therefore, for 4:2:2 and 4:4:4 chroma formats, CC-RIF 2110 is not applied to samples in the 3rd and 4th lines above the CTU boundary.

[0206] In one example, CC-RIF 2110 may be performed by using the same diamond filter as the CC-ALF described above as a linear filter. To this end, for each chroma sample, a 3×4 sized diamond high pass filter (such as Fig.13 The example in ( ) can be applied to luma samples to generate correction values. To generate correction values ​​for each chroma sample, the example uses luma samples that have been SAO filtered corresponding to each chroma sample position. Thus, the use of CC-RIF 2110 can be replaced using the same filter shape and size as conventional CC-ALF.

[0207] As another example, CC-RIF 2110 may be performed by using a finite impulse response (FIR) filter. Specifically, an interpolation filter may be utilized to generate correction values ​​for chroma samples from input luma samples, which are applied to the generation of a prediction block during inter prediction.

[0208] As another example, CC-RIF 2110 upsamples the chrominance component by using a neural network super-resolution (NNSR) technique that performs upsampling based on a neural network. Specifically, a transposed convolution may be used, or a convolution process may be performed after upsampling. By using a transposed convolution, i.e., the inverse of the mathematical operation used in convolution, the image may be upsampled, and the kernel used may be updated by using learning. Figure 7 The deconvolution layer shown in implements transposed convolution. In addition, the network structure and component techniques for super-resolution processing (such as SISR and EDSR mentioned above) can be used as NNSR.

[0209] Fig. 22 is a diagram illustrating the application of neural network-based CC-RIF according to at least one embodiment of the present disclosure.

[0210] like Fig. 22 As shown in , when scaling the chrominance component, the video decoding device can use the luminance component and the chrominance component together to improve the encoding efficiency. Fig. 22 In the example, the operation ⊙ represents a cascade.

[0211] Using the above-described super-resolution processing techniques, a network can be constructed for each of the luma and chroma components. In this case, the luma samples are input to the upsampling network for the chroma samples, i.e., CC-RIF 2110.

[0212] The upsampling network of the luminance component (i.e., luminance RIF (RIF luminance)) uses a low-resolution reconstructed frame Y LR rec As input, it generates an upsampled high-resolution reconstructed frame Y HR rec As output, such as Fig.23 as shown in the example.

[0213] Similar to the brightness component network, Fig.24 As shown in the example of , a chroma component upsampling network is constructed. Here, the upsampling scaling follows the scaling of the existing brightness RIF. In order to obtain the correlation between the brightness component and the chroma component, the brightness component Y LR recThe feature map extracted from the luminance component and the feature map extracted from the chrominance component are concatenated, and then the concatenated feature map is input into the convolutional layer to generate an upsampled high-resolution reconstructed frame Ci HR rec .

[0214] As described above, the super-resolution model can be used as an upsampling network. In this case, both the luma component upsampling network and the chroma component upsampling network can additionally use the luma component Y of the predicted frame, such as LR pred , quantization parameter map, etc. as input. Here, the quantization parameter map has the same size as the input frame and represents a map in which each pixel value for each CU is a quantization parameter. If there are more than two inputs, all inputs are matched to the same resolution as the chroma components, and the matched inputs are concatenated and applied to the convolution layer. For example, the luminance component Y LR rec The downsampling can be performed to the same resolution as the chroma components by using a convolutional layer with a stride of 2 for the 4:2:0 format. Alternatively, the luma component can be downsampled to the same resolution as the chroma components by using a max pooling stride of 2.

[0215] Since the two chroma components may have different characteristics, a super-resolution model may be trained for each of the Cb and Cr components. Alternatively, a network with the same structure and parameters may be used with only the chroma component inputs Cb and Cr changed.

[0216] Alternatively, whether to apply CC-RIF 2110 may be determined at the level of a sequence, a picture, a sub-picture, a slice, a tile, and / or a CTU.

[0217] In one example, CC-RIF 2110 depends on ALF 566 for the relevant applicable unit level. For example, if ALF 566 is used at the relevant applicable unit level, CC-RIF 2110 may also be used. For example, for CC-RIF 2110 used at the slice level, ALF 566 is also required to be used at the slice.

[0218] When a FIR filter is used as CC-RIF 2110, the video decoding device may use one of multiple FIR filters based on the scaling value. The corresponding set of filter coefficients may be stored in an adaptation parameter set (APS) and signaled by the video encoding device.

[0219] When using CC-RIF 2110 based on a neural network, an index indicating one of a plurality of candidate super-resolution neural networks may be signaled. The index may indicate a parameter set of the corresponding neural network. Based on the signaled index, the video decoding device sets the neural network for super-resolution processing. The candidate neural network may be pre-trained based on quantization parameters, scaling values, etc. The video encoding device and the video decoding device may store and use the generally structured neural network and related parameters.

[0220] Alternatively, to perform adaptive upsampling on video content, the video encoding device may store a parameter set of the neural network in a bitstream and then signal the parameter set to the video decoding device. In this case, the parameter set may be transmitted when stored in the APS.

[0221] Reference now Fig.25 and Fig.26 , describes a method for upsampling reconstructed frames based on CC-RIF.

[0222] Fig.25 The present invention is a flowchart of a method for upsampling a reconstructed frame performed by a video encoding device according to at least one embodiment of the present disclosure.

[0223] The video encoding apparatus obtains a scaling value (S2500).

[0224] Here, the scaling value represents the resolution difference between the reconstructed frame and the original frame. Further, the reconstructed frame) as the reconstructed frame of the original frame) includes luminance samples and chrominance samples. In addition, the reconstructed frame can be pre-reconstructed according to the inter-frame prediction of the video encoding device.

[0225] On the other hand, if the original frame is downsampled, the scaling value may be set to 2.

[0226] Alternatively, in terms of rate-distortion optimization, the video encoding device may determine a scaling value, may encode an index indicating the scaling value, and may communicate the encoded index to the video decoding device.

[0227] The video encoding apparatus obtains a reconstructed frame ( S2502 ).

[0228] The video encoding apparatus checks the scaling value (S2504).

[0229] If the resolution of the reconstructed frame based on the scaling value is smaller than the resolution of the original frame (Yes in S2504 ), the video encoding apparatus performs the following steps ( S2506 and S2512 ).

[0230] The video encoding apparatus inputs the luma sample and the chroma sample into CC-RIF to generate an upsampled chroma frame (S2506). Here, CC-RIF replaces one of the chroma component filters including the deblocking filter, the SAO filter, and the adaptive loop filter, or CC-RIF is inserted between the filters. The upsampled chroma frame has the resolution of the original chroma frame.

[0231] In one example, CC-RIF may be a linear filter. In this case, the video encoding device inputs the chroma samples to the chroma resampling in-loop filter (C-RIF) to generate upsampled chroma samples. The video encoding device inputs the luma samples to the CC-RIF to generate correction values ​​for the upsampled chroma samples. By summing the upsampled chroma samples with the aforementioned correction values, the video encoding device may generate an upsampled chroma frame. The linear filter may be a diamond high-pass filter of size 3×4 used by CC-ALF.

[0232] As another example, the video encoding apparatus may use a finite impulse response (FIR) filter as the CC-RIF. Here, the FIR filter may be an interpolation filter applied to the generation of a prediction block during inter prediction.

[0233] Further, the video encoding device may utilize one of a plurality of linear filters / FIR filters based on the scaling value.The video encoding device may signal a corresponding set of filter coefficients when the video encoding device stores the filter coefficients in an adaptive parameter set (APS).

[0234] As mentioned above, when CC-RIF is applied to 4:2:2 or 4:4:4 chroma formats, due to the difference in the position of the virtual boundary, the luma line buffer and the chroma line buffer are not aligned with each other for the third and fourth rows above the CTU boundary. Therefore, for 4:2:2 and 4:4:4 chroma formats, CC-RIF may not be applied to samples in the third and fourth rows above the CTU boundary.

[0235] As another example, CC-RIF may be a chroma upsampling network based on deep learning. In this case, the video encoding device may concatenate luma samples and chroma samples to generate concatenated samples, and then the concatenated samples may be input to the chroma upsampling network to generate an upsampled chroma frame.

[0236] The chroma upsampling network may be implemented as a super-resolution model and may additionally take as input a prediction frame and a quantization parameter map for the luma component. Further, the chroma upsampling network may include a separate super-resolution model for each of the Cb and Cr chroma components.

[0237] In terms of rate-distortion optimization, the video encoding device determines an index indicating one of the plurality of neural network candidates as a chroma upsampling network. Subsequently, the video encoding device may encode the determined index. At this time, each neural network candidate may be pre-trained based on a quantization parameter and a scaling value.

[0238] Alternatively, the video encoding device may encode a parameter set of a neural network selected as a chroma upsampling network from a plurality of neural network candidates. In this case, the parameter set may be signaled when stored in the APS.

[0239] The video encoding apparatus inputs the luma samples to a luma resampling in-loop filter (luma RIF) to generate an up-sampled luma frame (S2508).

[0240] The luma RIF may be a luma upsampling network based on deep learning. In this case, the video encoding device may input luma samples to the luma upsampling network to generate an upsampled luma frame. The luma upsampling network may be implemented as a super-resolution model, and may further use a predicted frame and a quantization parameter map of the luma component as input.

[0241] The video encoding device determines a flag indicating whether CC-RIF is used (S2510). The video encoding device may determine the flag indicating whether CC-RIF is used based on whether the original frame is downsampled based on the scaling value. That is, if the original frame is downsampled, the video encoding device may set the flag to true.

[0242] The video encoding apparatus encodes a flag indicating whether CC-RIF is to be used (S2512).

[0243] On the other hand, if the resolution of the reconstructed frame is the same as the resolution of the original frame (No in S2504), the video encoding apparatus performs conventional in-loop filtering, wherein CC-RIF is excluded (S2520).

[0244] In this case, the video encoding device may set the flag indicating whether CC-RIF is used to false, and then encode the set flag. Alternatively, the video encoding device may not transmit the flag indicating whether CC-RIF is used. If the flag indicating whether CC-RIF is used is not transmitted, the video decoding device may infer that the value of the flag is false.

[0245] Fig.26 The present invention is a flowchart of a method for upsampling a reconstructed frame performed by a video decoding device according to at least one embodiment of the present disclosure.

[0246] The video decoding apparatus obtains a reconstructed frame (S2600). Here, the reconstructed frame (which is a reconstruction of the original frame) includes luminance samples and chrominance samples. In addition, the reconstructed frame can be pre-reconstructed according to the inter-frame prediction of the video decoding apparatus.

[0247] The video decoding apparatus decodes a flag indicating whether CC-RIF is enabled or disabled from a bitstream (S2602), and checks the flag (S2604).

[0248] On the other hand, if the flag indicating whether CC-RIF is enabled or disabled is not transmitted, the video decoding device may infer that the value of the flag is false.

[0249] If the flag indicating whether CC-RIF is enabled is true ("Yes" in S2604), the video decoding apparatus performs the following steps (step S2606 and following steps).

[0250] The video decoding apparatus obtains a scaling value (S2606). Here, the scaling value represents a resolution difference between the reconstructed frame and the original frame.

[0251] If the flag indicating whether CC-RIF is enabled is true, the scaling value may be set to 2.

[0252] Alternatively, the video decoding device may decode an index indicating a scaling value from a bitstream and set the scaling value based on the index.

[0253] In order to increase the resolution of the reconstructed frame by scaling the value, the video decoding device may input the luminance sample and the chrominance sample into the CC-RIF to generate an upsampled chrominance frame (S2608). Here, the CC-RIF replaces one of the chrominance component filters including the deblocking filter, the SAO filter, and the adaptive loop filter, or the CC-RIF is inserted between the filters. The upsampled chrominance frame has the resolution of the original frame.

[0254] In one example, CC-RIF may be a linear filter. The video decoding device inputs the chroma samples to the chroma resampling in-loop filter (C-RIF) to generate upsampled chroma samples. The video decoding device inputs the luminance samples to the CC-RIF to generate correction values ​​for the upsampled chroma samples. By summing the upsampled chroma samples with the aforementioned correction values, the video decoding device may generate an upsampled chroma frame. The linear filter may be a diamond high-pass filter of size 3×4 used by CC-ALF.

[0255] As another example, the video decoding apparatus may use an FIR filter as the CC-RIF. Here, the FIR filter may be an interpolation filter applied to generation of a prediction block during inter-frame prediction.

[0256] Further, the video decoding device may utilize one of a plurality of linear filters / FIR filters based on the scaling value.The video decoding device may decode a corresponding set of filter coefficients from an adaptive parameter set (APS) in the bitstream.

[0257] As mentioned above, when CC-RIF is applied to 4:2:2 or 4:4:4 chroma formats, due to the difference in the position of the virtual boundary, the luma line buffer and the chroma line buffer are not aligned with each other for the third and fourth rows above the CTU boundary. Therefore, for 4:2:2 and 4:4:4 chroma formats, CC-RIF may not be applied to samples in the third and fourth rows above the CTU boundary.

[0258] As another example, CC-RIF may be a chroma upsampling network based on deep learning. In this case, the video decoding device may concatenate luma samples and chroma samples to generate concatenated samples, and then the concatenated samples may be input to the chroma upsampling network to generate an upsampled chroma frame.

[0259] The chroma upsampling network may be implemented as a super-resolution model and may additionally take as input a prediction frame and a quantization parameter map for the luma component. Further, the chroma upsampling network may include a separate super-resolution model for each of the Cb and Cr chroma components.

[0260] The video decoding device may decode an index indicating one of a plurality of neural network candidates from a bitstream, and then may establish a chroma upsampling network based on the index. Each neural network candidate may be pre-trained based on a quantization parameter and a scaling value.

[0261] Alternatively, the video decoding device may decode the parameter set of the chroma upsampling network from the bitstream. In this case, the parameter set may be signaled when stored in the APS.

[0262] The video decoding apparatus inputs luma samples to a luma resampling in-loop filter (luma RIF) to generate an up-sampled luma frame (S2610).

[0263] The luma RIF may be a luma upsampling network based on deep learning. In this case, the video decoding device may input luma samples into the luma upsampling network to generate an upsampled luma frame. The luma upsampling network may be implemented as a super-resolution model, and may also take a predicted frame and a quantization parameter map of the luma component as input.

[0264] On the other hand, if the flag indicating whether CC-RIF is enabled is false (No in S2604), the video decoding apparatus performs conventional in-loop filtering in which CC-RIF is excluded (S2620).

[0265] Although the steps in each flowchart are described as being performed sequentially, these steps are merely illustrative of the technical concepts of some embodiments of the present disclosure. Therefore, a person of ordinary skill in the art to which the present disclosure belongs can perform the steps by changing the order described in each figure or by performing two or more steps in parallel. Therefore, the steps in each flowchart are not limited to the time series order shown.

[0266] It should be understood that the above description presents illustrative embodiments that can be implemented in various other ways. The functions described in some embodiments can be implemented by hardware, software, firmware and / or a combination thereof. It should also be understood that the functional components described in this disclosure are marked with "... units" in order to strongly emphasize the possibility of their independent implementation.

[0267] At the same time, the various methods or functions described in some embodiments may be implemented as instructions stored in a non-transitory recording medium that can be read and executed by one or more processors. For example, the non-transitory recording medium may include various types of recording devices, in which data is stored in a form readable by a computer system. For example, the non-transitory recording medium may include a storage medium such as an erasable programmable read-only memory (EPROM), a flash drive, an optical drive, a magnetic hard drive, and a solid-state drive (SSD).

[0268] Although the embodiments of the present disclosure have been described for illustrative purposes, however, those of ordinary skill in the art to which the present disclosure belongs should recognize that various modifications, additions, and substitutions are possible without departing from the concept and scope of the present disclosure. Therefore, for the sake of brevity and clarity, the embodiments of the present disclosure have been described. The scope of the technical concept of the embodiments of the present disclosure is not limited by the diagrams. Therefore, those of ordinary skill in the art to which the present disclosure belongs should understand that the scope of the present disclosure should not be limited by the embodiments explicitly described above, but by the claims and their equivalents.

[0269] (reference number)

[0270] 180 Loop filter unit

[0271] 155 Entropy Encoder

[0272] 510 Entropy Decoder

[0273] 560 Loop Filter Unit

[0274] 1910 Chroma component resampling in-loop filter

[0275] 2110 Cross-component resampling in-loop filter

[0276] CROSS-REFERENCE TO RELATED APPLICATIONS

[0277] This application claims priority to and the benefit of Korean Patent Application No. 10-2022-0046880, filed on April 15, 2022, and Korean Patent Application No. 10-2023-0037117, filed on March 22, 2023, the entire contents of each of which are incorporated herein by reference.

Claims

1. A method for upsampling a reconstructed frame performed by a video decoding device, the method include: Obtaining the reconstructed frame, wherein the reconstructed frame is a reconstruction of the original frame, and the reconstructed frame includes luma samples and chroma samples; obtaining a scaling value representing a resolution difference between the reconstructed frame and the original frame; and generating an upsampled chroma frame by inputting the luma samples and the chroma samples into a cross-component resampling in-loop filter (CC-RIF) based on the scaled value, The CC-RIF replaces one of the filters for the chrominance component, or the CC-RIF is inserted between the filters, wherein the filters include a deblocking filter, a sample adaptive offset filter (SAO filter), and an adaptive loop filter.

2. The method according to claim 1, further comprising: include: decoding from a bitstream a flag indicating whether the CC-RIF is enabled; as well as When the flag is true, the method further includes obtaining the scaling value.

3. The method according to claim 1, further comprising: include: Based on the scaling value, an upsampled luma frame is generated by inputting the luma samples into a luma resampling in-loop filter (luma RIF).

4. The method according to claim 1, in, Generating the upsampled chroma frame comprises: generating upsampled chroma samples by inputting the chroma samples into a chroma resampling in-loop filter (C-RIF); generating correction values ​​for the upsampled chroma samples by inputting the luma samples into the CC-RIF; and The upsampled chroma samples and the correction values ​​are summed.

5. The method according to claim 4, in, The CC-RIF generates a correction value corresponding to a correlation between chroma samples and luma samples at corresponding positions based on a finite impulse response filter (FIR filter), which is an interpolation filter applied to generate a prediction block during inter-frame prediction.

6. The method according to claim 4, in, The CC-RIF generates a correction value corresponding to a correlation between chroma samples and luma samples at corresponding positions based on a linear filter, wherein the linear filter is a diamond high-pass filter of size 3×4.

7. The method according to claim 1, in, The CC-RIF does not apply to samples in the third and fourth rows above a Coding Tree Unit (CTU) boundary for 4:2:2 and 4:4:4 chroma formats.

8. The method according to claim 1, in, Generating an upsampled chroma frame involves: generating a cascaded sample by cascading the luma sample and the chroma sample; and The upsampled chroma frame is generated by inputting the cascaded samples into a chroma upsampling network, wherein the chroma upsampling network is based on deep learning and used as the CC-RIF.

9. The method according to claim 8, in, The chroma upsampling network is implemented as a super-resolution model, and the chroma upsampling network also takes the predicted frame and quantization parameter map of the luma component as input.

10. The method according to claim 8, further comprising: include: decoding an index indicating one of a plurality of neural network candidates from a bitstream; as well as Setting the chroma upsampling network based on the index, Wherein, each of the neural network candidates is pre-trained based on the scaling value.

11. A method for upsampling a reconstructed frame performed by a video encoding device, the method include: Obtaining a scaling value representing a resolution difference between the reconstructed frame and the original frame, the reconstructed frame being a reconstruction of the original frame, and the reconstructed frame comprising luma samples and chroma samples; Obtaining the reconstructed frame; as well as When the reconstructed frame has a resolution based on the scaling value and is less than a resolution of the original frame, generating an upsampled chroma frame by inputting the luma samples and the chroma samples into a cross-component resampling in-loop filter (CC-RIF), The CC-RIF replaces one of the filters for the chrominance component, or the CC-RIF is inserted between the filters, wherein the filters include a deblocking filter, a sample adaptive offset filter (SAO filter), and an adaptive loop filter.

12. The method according to claim 12, further comprising: include: determining a flag indicating whether the CC-RIF is enabled according to whether the original frame is downsampled based on the scaling value; as well as The flag is encoded.

13. The method according to claim 12, in, Determining the flag includes: The flag is set to true when the original frame is downsampled.

14. The method according to claim 11, further comprising: include: When the reconstructed frame has a resolution based on the scaling value and is less than a resolution of the original frame, an upsampled luma frame is generated by inputting the luma samples into a luma resampling in-loop filter (luma RIF).

15. The video encoding method according to claim 11, in, Generating an upsampled chroma frame involves: generating a cascaded sample by cascading the luma sample and the chroma sample; and The upsampled chroma frame is generated by inputting the cascaded samples into a chroma upsampling network, wherein the chroma upsampling network is based on deep learning and used as the CC-RIF.

16. The video encoding method according to claim 15, further comprising: include: determining an index indicative of one of a plurality of neural network candidates for the chroma upsampling network; as well as encoding the index, Wherein, each of the neural network candidates is pre-trained based on the scaling value.

17. A computer-readable recording medium storing a bit stream generated by a video encoding method, the video encoding method include: Obtaining a scaling value representing a resolution difference between a reconstructed frame and an original frame, wherein the reconstructed frame is a reconstruction of the original frame and the reconstructed frame includes luma samples and chroma samples; Obtaining the reconstructed frame; as well as When the reconstructed frame has a resolution based on the scaling value and is less than a resolution of the original frame, generating an upsampled chroma frame by inputting the luma samples and the chroma samples into a cross-component resampling in-loop filter (CC-RIF), The CC-RIF replaces one of the filters for the chrominance component, or the CC-RIF is inserted between the filters, wherein the filters include a deblocking filter, a sample adaptive offset filter (SAO filter), and an adaptive loop filter.