Deepfake detection method and device

Abstract

A deepfake detection method and device are provided. The deepfake detection method may comprise the steps of: receiving a bitstream of an image; obtaining syntax information from the bitstream; calculating a unreliability measurement value on the basis of the syntax information; and determining whether the image is a deepfake on the basis of the unreliability measurement value, wherein the syntax information may be at least one of a quantization parameter, a transform coefficient, a bit rate, a motion vector, a reference picture index, or a boundary strength of a deblocking filter.

Description

Deepfake detection method and device

[0001] The present disclosure relates to a method and device for detecting deepfakes. Specifically, the present disclosure relates to a method and device for detecting deepfakes based on syntactic element information in a bitstream.

[0002] Deepfake refers to a technology that uses deep learning technology to create manipulated videos by superimposing other images on top of video. Deepfake can be used in industries such as film, television, and television shows to create creative works by recreating actors' performances or the appearance of specific individuals. Recently, virtual YouTubers have emerged, livestreaming with a virtual face that resembles a real person's body and voice but lacks a real face.

[0003] As deepfakes are used across a wide range of fields, concerns about the technology are growing. While video manipulation in the past required significant time and expense, the emergence of deepfakes makes it easy for anyone to manipulate videos, and this can be exploited.

[0004] The present disclosure aims to provide a deepfake detection method and device to solve the above problems.

[0005] A deepfake detection method according to one embodiment of the present disclosure includes the steps of receiving a bitstream of an image, obtaining syntax information from the bitstream, calculating an unreliability measurement value based on the syntax information, and determining whether the image is a deepfake based on the unreliability measurement value, wherein the syntax information may be at least one of a quantization parameter, a transform coefficient, a bit rate, a motion vector, a reference picture index, or a boundary strength of a deblocking filter.

[0006] In the above deepfake detection method, the step of obtaining the syntax information may obtain syntax information of a region of interest within the current picture.

[0007] In the above deepfake detection method, the region of interest can be determined as a region in which a motion vector exceeding a predefined value exists.

[0008] In the above deepfake detection method, the region of interest can be determined as a region in which a transformation coefficient greater than a predefined value exists.

[0009] In the above deepfake detection method, the region of interest can be determined based on object detection.

[0010] In the above deepfake detection method, whether the video is a deepfake can be determined based on multiple unreliability measurement values ​​calculated based on multiple pieces of syntax information.

[0011] In the above deepfake detection method, whether the video is a deepfake can be determined based on a distrust score derived by applying a weight to each of a plurality of distrust measurement values.

[0012] In the above deepfake detection method, the distrust score can be derived by normalizing each of the plurality of distrust measurement values ​​and applying the weight.

[0013] In the above deepfake detection method, whether the image is a deepfake can be determined based on the unreliability measurement value and spatial domain information.

[0014] In the above deepfake detection method, the spatial domain information may be at least one of illumination change information, sharpness information, contrast information, or blinking information.

[0015] In the above deepfake detection method, whether the image is a deepfake can be determined based on an unreliability score derived by applying a weight to each of the unreliability measurement value and the spatial domain information.

[0016] A deepfake detection device according to one embodiment of the present disclosure includes a receiving unit that receives a bitstream of an image, and a detection unit that obtains syntax information from the bitstream, calculates an unreliability measurement value based on the syntax information, and determines whether the image is a deepfake based on the unreliability measurement value, wherein the syntax information may be at least one of a quantization parameter, a transform coefficient, a bit rate, a motion vector, a reference picture index, or a deblocking filter boundary strength.

[0017] The features briefly summarized above regarding the present disclosure are merely exemplary aspects of the detailed description of the present disclosure that follows and do not limit the scope of the present disclosure.

[0018] According to the present disclosure, a deepfake detection method and device with improved deepfake detection efficiency can be provided.

[0019] The effects that can be obtained from the present disclosure are not limited to the effects mentioned above, and other effects that are not mentioned will be clearly understood by a person having ordinary skill in the art to which the present disclosure pertains from the description below.

[0020] FIG. 1 is a block diagram showing a configuration according to one embodiment of an encoding device to which the present disclosure is applied.

[0021] FIG. 2 is a block diagram showing a configuration according to one embodiment of a decryption device to which the present disclosure is applied.

[0022] FIG. 3 is a diagram schematically illustrating a video coding system to which the present disclosure can be applied.

[0023] FIG. 4 is a flowchart illustrating a deepfake detection method according to an embodiment of the present disclosure.

[0024] FIGS. 5 and 6 are drawings for explaining conversion coefficients for calculating unreliable measurement values ​​according to one embodiment of the present disclosure.

[0025] FIG. 7 is a diagram illustrating a deepfake detection method using multiple syntax elements according to one embodiment of the present disclosure.

[0026] FIG. 8 is a diagram illustrating a deepfake detection method using an unreliability score according to an embodiment of the present disclosure.

[0027] FIG. 9 is a diagram illustrating a deepfake detection method based on a normalized unreliability measurement value according to one embodiment of the present disclosure.

[0028] FIG. 10 is a flowchart illustrating a deepfake detection method according to an embodiment of the present disclosure.

[0029] FIG. 11 is a block diagram illustrating a deepfake detection device according to an embodiment of the present disclosure.

[0030] FIG. 12 is a diagram exemplifying a content streaming system to which an embodiment according to the present disclosure can be applied.

[0031] A deepfake detection method according to one embodiment of the present disclosure includes the steps of receiving a bitstream of an image, obtaining syntax information from the bitstream, calculating an unreliability measurement value based on the syntax information, and determining whether the image is a deepfake based on the unreliability measurement value, wherein the syntax information may be at least one of a quantization parameter, a transform coefficient, a bit rate, a motion vector, a reference picture index, or a boundary strength of a deblocking filter.

[0032] The present disclosure is susceptible to various modifications and embodiments. Therefore, specific embodiments are illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the present disclosure to specific embodiments, but rather to encompass all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosure. In the drawings, similar reference numerals designate the same or similar functions throughout. The shapes and sizes of elements in the drawings may be provided by way of example only for clarity. The detailed description of the exemplary embodiments described below refers to the accompanying drawings, which illustrate specific embodiments by way of example. These embodiments are described in sufficient detail to enable those skilled in the art to practice the embodiments. It should be understood that the various embodiments, while different from each other, are not necessarily mutually exclusive. For example, specific shapes, structures, and characteristics described herein may be implemented in other embodiments without departing from the spirit and scope of the present disclosure. Furthermore, it should be understood that the positions or arrangements of individual components within each disclosed embodiment may be modified without departing from the spirit and scope of the embodiment. Accordingly, the detailed description set forth below is not intended to be taken in a limiting sense, and the scope of the illustrative embodiments, if properly described, is defined only by the appended claims, along with the full scope equivalents to which such claims are entitled.

[0033] While terms such as "first" and "second" may be used herein to describe various components, these components should not be limited by these terms. These terms are used solely to distinguish one component from another. For example, without departing from the scope of the present disclosure, a first component may be referred to as a "second component," and similarly, a second component may also be referred to as a "first component." The term "and / or" includes a combination of multiple related items described herein or any of multiple related items described herein.

[0034] The components shown in the embodiments of the present disclosure are independently depicted to represent different characteristic functions, and do not imply that each component is composed of separate hardware or a single software component. That is, each component is listed and included as a separate component for convenience of explanation, and at least two components among each component may be combined to form a single component, or a single component may be divided into multiple components to perform a function, and such integrated and separate embodiments of each component are also included in the scope of the present disclosure as long as they do not deviate from the essence of the present disclosure.

[0035] The terminology used in this disclosure is merely used to describe specific embodiments and is not intended to limit the present disclosure. The singular expression includes the plural expression unless the context clearly dictates otherwise. In addition, some components of the present disclosure may not be essential components that perform essential functions in the present disclosure and may be optional components merely for performance enhancement. The present disclosure may be implemented by including only components essential to implementing the essence of the present disclosure, excluding components used solely for performance enhancement, and a structure including only essential components, excluding optional components used solely for performance enhancement, is also within the scope of the present disclosure.

[0036] In embodiments, the term "at least one" may mean one of a number greater than or equal to 1, such as 1, 2, 3, and 4. In embodiments, the term "a plurality of" may mean one of a number greater than or equal to 2, such as 2, 3, and 4.

[0037] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. In describing the embodiments of this specification, if a detailed description of a related known configuration or function is judged to obscure the gist of this specification, the detailed description will be omitted. The same reference numerals will be used for identical components in the drawings, and duplicate descriptions of the same components will be omitted.

[0038] Glossary of Terms

[0039] Hereinafter, “video” may mean a single picture constituting a video, or may refer to the video itself. For example, “encoding and / or decoding of a video” may mean “encoding and / or decoding of a video,” or may mean “encoding and / or decoding of one of the videos constituting the video.”

[0040] Hereinafter, the terms "video" and "movie" may be used interchangeably and have the same meaning. Furthermore, the target image may be an encoding target image, which is the target of encoding, and / or a decoding target image, which is the target of decoding. Furthermore, the target image may be an input image input to an encoding device, or an input image input to a decoding device. Here, the target image may have the same meaning as the current image.

[0041] Hereinafter, the terms encoder and image encoding device may be used interchangeably and have the same meaning.

[0042] Hereinafter, the terms decoder and image decoding device may be used interchangeably and have the same meaning.

[0043] Hereinafter, “image”, “picture”, “frame” and “screen” may be used with the same meaning and may be used interchangeably.

[0044] Hereinafter, the term "target block" may refer to an encoding target block, which is the target of encoding, and / or a decoding target block, which is the target of decoding. Furthermore, the target block may refer to a current block, which is the target of current encoding and / or decoding. For example, the terms "target block" and "current block" may be used interchangeably and have the same meaning.

[0045] Hereinafter, "block" and "unit" may be used with the same meaning and may be used interchangeably. In addition, "unit" may mean including a luminance component block and a corresponding chroma component block to distinguish it from a block. For example, a coding tree unit (CTU) may be composed of one luma component (Y) coding tree block (CTB) and two chroma component (Cb, Cr) coding tree blocks associated with it.

[0046] Hereinafter, the terms “sample,” “pixel,” and “pixel” may be used interchangeably and have the same meaning. Here, a sample may represent a basic unit that constitutes a block.

[0047] Hereinafter, “inter” and “between screens” may be used interchangeably and have the same meaning.

[0048] Hereinafter, “intra” and “within screen” may be used interchangeably and have the same meaning.

[0049]

[0050] FIG. 1 is a block diagram showing a configuration according to one embodiment of an encoding device to which the present disclosure is applied.

[0051] The encoding device (100) may be an encoder, a video encoding device, or an image encoding device. A video may include one or more images. The encoding device (100) may sequentially encode one or more images.

[0052] Referring to FIG. 1, the encoding device (100) may include an image segmentation unit (110), an intra prediction unit (120), a motion prediction unit (121), a motion compensation unit (122), a switch (115), a subtractor (113), a transformation unit (130), a quantization unit (140), an entropy encoding unit (150), an inverse quantization unit (160), an inverse transformation unit (170), an adder (117), a filter unit (180), and a reference picture buffer (190).

[0053] Additionally, the encoding device (100) can generate a bitstream including encoded information through encoding an input image and output the generated bitstream. The generated bitstream can be stored in a computer-readable recording medium or can be streamed via a wired / wireless transmission medium.

[0054] The video segmentation unit (110) can segment the input video into various forms to increase the efficiency of video encoding / decoding. That is, the input video is composed of multiple pictures, and one picture can be hierarchically segmented and processed for compression efficiency, parallel processing, etc. For example, one picture can be segmented into one or more tiles or slices, which can then be segmented into multiple Coding Tree Units (CTUs). Alternatively, one picture can first be segmented into multiple sub-pictures defined as groups of rectangular slices, and each sub-picture can then be segmented into the tiles / slices. Here, the sub-pictures can be utilized to support the function of partially independently encoding / decoding and transmitting the picture. Since multiple sub-pictures can each be individually restored, there is an advantage of easy editing in applications that configure multi-channel input into a single picture. In addition, tiles can be segmented horizontally to generate bricks. Here, a brick can be utilized as the basic unit of intra-picture parallel processing. In addition, one CTU can be recursively split into a quadtree (QT), and the terminal node of the split can be defined as a coding unit (CU). The CU can be split into a prediction unit (PU) and a transformation unit (TU), and prediction and splitting can be performed. Meanwhile, the CU can be utilized as a prediction unit and / or a transformation unit itself. Here, for flexible splitting, each CTU can be recursively split into a multi-type tree (MTT) as well as a quadtree (QT). Splitting of a CTU into a multi-type tree can start from the terminal node of a QT, and the MTT can be composed of a binary tree (BT) and a triple tree (TT).For example, the MTT structure can be divided into vertical binary split mode (SPLIT_BT_VER), horizontal binary split mode (SPLIT_BT_HOR), vertical ternary split mode (SPLIT_TT_VER), and horizontal ternary split mode (SPLIT_TT_HOR). In addition, the minimum block size (MinQTSize) of the quad tree of the luminance block during splitting can be set to 16x16, the maximum block size (MaxBtSize) of the binary tree can be set to 128x128, and the maximum block size (MaxTtSize) of the triple tree can be set to 64x64. In addition, the minimum block size (MinBtSize) of the binary tree and the minimum block size (MinTtSize) of the triple tree can be set to 4x4, and the maximum depth (MaxMttDepth) of the multi-type tree can be set to 4. Additionally, to improve the encoding efficiency of the I slice, a dual tree can be applied that uses different CTU partition structures for luminance and chrominance components. On the other hand, in the P and B slices, the luminance and chrominance CTBs (Coding Tree Blocks) within the CTU can be partitioned into a single tree that shares the coding tree structure.

[0055] The encoding device (100) may perform encoding on the input image in intra mode and / or inter mode. Alternatively, the encoding device (100) may perform encoding on the input image in a third mode (e.g., IBC mode, Palette mode, etc.) other than the intra mode and inter mode. However, if the third mode has functional characteristics similar to the intra mode or inter mode, it may be classified as intra mode or inter mode for convenience of explanation. In the present disclosure, the third mode will be classified and described separately only when a specific description is required.

[0056] When the intra mode is used as the prediction mode, the switch (115) can be switched to intra, and when the inter mode is used as the prediction mode, the switch (115) can be switched to inter. Here, the intra mode can mean an intra-screen prediction mode, and the inter mode can mean an inter-screen prediction mode. The encoding device (100) can generate a prediction block for an input block of an input image. In addition, after the prediction block is generated, the encoding device (100) can encode a residual block using a residual of the input block and the prediction block. The input image can be referred to as a current image that is currently a target of encoding. The input block can be referred to as a current block that is currently a target of encoding or an encoding target block.

[0057] When the prediction mode is intra mode, the intra prediction unit (120) can use samples of blocks already encoded / decoded around the current block as reference samples. The intra prediction unit (120) can perform spatial prediction on the current block using the reference samples, and can generate prediction samples for the input block through spatial prediction. Here, intra prediction can mean prediction within the screen.

[0058] As an intra prediction method, non-directional prediction modes such as DC mode and Planar mode, as well as directional prediction modes (e.g., 65 directions) can be applied. Here, the intra prediction method can be expressed as an intra prediction mode or an intra-screen prediction mode.

[0059] When the prediction mode is inter mode, the motion prediction unit (121) can search for an area that best matches the input block from the reference image during the motion prediction process and derive a motion vector using the searched area. At this time, the area can be used as a search area. The reference image can be stored in the reference picture buffer (190). Here, when encoding / decoding for the reference image is processed, it can be stored in the reference picture buffer (190).

[0060] The motion compensation unit (122) can generate a prediction block for the current block by performing motion compensation using a motion vector. Here, inter prediction may mean inter-screen prediction or motion compensation.

[0061] The above motion prediction unit (121) and motion compensation unit (122) can generate a prediction block by applying an interpolation filter to a portion of an area within a reference image when the value of the motion vector does not have an integer value. In order to perform inter-screen prediction or motion compensation, it is possible to determine whether the motion prediction and motion compensation method of the prediction unit included in the corresponding encoding unit is one of Skip Mode, Merge Mode, Advanced Motion Vector Prediction (AMVP) mode, and Intra Block Copy (IBC) mode based on the encoding unit, and perform inter-screen prediction or motion compensation according to each mode.

[0062] In addition, based on the above inter-screen prediction method, the AFFINE mode of sub-PU based prediction, the SbTMVP (Subblock-based Temporal Motion Vector Prediction) mode, and the MMVD (Merge with MVD) mode and the GPM (Geometric Partitioning Mode) mode of PU based prediction can be applied. In addition, in order to improve the performance of each mode, the HMVP (History based MVP), the PAMVP (Pairwise Average MVP), the CIIP (Combined Intra / Inter Prediction), the AMVR (Adaptive Motion Vector Resolution), the BDOF (Bi-Directional Optical-Flow), the BCW (Bi-predictive with CU Weights), the LIC (Local Illumination Compensation), the TM (Template Matching), and the OBMC (Overlapped Block Motion Compensation) can be applied.

[0063] Among these, AFFINE mode is a technology that is used in both AMVP and MERGE modes and also has high encoding efficiency. In the existing video coding standard, since MC (Motion Compensation) is performed by considering only the parallel translation of the block, there was a disadvantage in that it could not properly compensate for motions that occur in reality, such as zoom in / out and rotation. To supplement this, a 4-parameter affine motion model using two control point motion vectors (CPMV) and a 6-parameter affine motion model using three control point motion vectors can be applied to inter prediction. Here, CPMV is a vector representing the affine motion model of one of the upper left, upper right, and lower left of the current block.

[0064] The subtractor (113) can generate a residual block using the difference between the input block and the predicted block. The residual block may also be referred to as a residual signal. The residual signal may refer to the difference between the original signal and the predicted signal. Alternatively, the residual signal may be a signal generated by transforming, quantizing, or transforming and quantizing the difference between the original signal and the predicted signal. The residual block may be a residual signal in block units.

[0065] The transform unit (130) can perform a transform on the residual block to generate a transform coefficient and output the generated transform coefficient. Here, the transform coefficient may be a coefficient value generated by performing a transform on the residual block. When the transform skip mode is applied, the transform unit (130) may also skip the transform on the residual block.

[0066] Quantized levels can be generated by applying quantization to transform coefficients or residual signals. In the following embodiments, quantized levels may also be referred to as transform coefficients.

[0067] For example, a 4x4 luminance residual block generated through intra prediction can be transformed using a basis vector based on DST (Discrete Sine Transform), and the remaining residual blocks can be transformed using a basis vector based on DCT (Discrete Cosine Transform). In addition, through RQT (Residual Quad Tree) technology, the transform block is divided into a quad tree shape for one block, and after performing transformation and quantization on each transform block divided through RQT, a coded block flag (cbf) can be transmitted to increase encoding efficiency when all coefficients become 0.

[0068] Another alternative is to apply Multiple Transform Selection (MTS) technology, which selectively performs transformation using multiple transformation bases. That is, instead of dividing CUs into TUs via RQT, a Sub-block Transform (SBT) technology can perform a function similar to TU division. Specifically, SBT is applied only to inter-screen prediction blocks, and unlike RQT, it can divide the current block into ½ or ¼ blocks vertically or horizontally, and then perform transformation on only one of the blocks. For example, in a vertically divided block, the transformation can be performed on the leftmost or rightmost block, and in a horizontally divided block, the transformation can be performed on the topmost or bottommost block.

[0069] Additionally, LFNST (Low Frequency Non-Separable Transform), a secondary transform technique that further transforms the residual signal converted to the frequency domain through DCT or DST, can be applied. LFNST additionally performs a transform on the low-frequency region of 4x4 or 8x8 in the upper left, which allows the residual coefficients to be concentrated in the upper left.

[0070] The quantization unit (140) can generate a quantized level by quantizing a transform coefficient or residual signal according to a quantization parameter (QP), and can output the generated quantized level. At this time, the quantization unit (140) can quantize the transform coefficient using a quantization matrix.

[0071] For example, a quantizer with QP values ​​of 0 to 51 can be used. Alternatively, if the image size is larger and high encoding efficiency is required, a QP of 0 to 63 can be used. In addition, a Dependent Quantization (DQ) method that uses two quantizers instead of a single quantizer can be applied. DQ performs quantization using two quantizers (e.g., Q0 and Q1), but even without signaling information about the use of a specific quantizer, the quantizer to be used for the next transform coefficient can be selected based on the current state through a state transition model.

[0072] The entropy encoding unit (150) can generate a bitstream by performing entropy encoding according to a probability distribution on values ​​produced by the quantization unit (140) or coding parameter values ​​produced during the encoding process, and can output the bitstream. The entropy encoding unit (150) can perform entropy encoding on information about image samples and information for decoding the image. For example, the information for decoding the image can include syntax elements, etc.

[0073] When entropy encoding is applied, a small number of bits are allocated to symbols with a high occurrence probability, and a large number of bits are allocated to symbols with a low occurrence probability, thereby representing the symbols, whereby the size of the bit string for the symbols to be encoded can be reduced. The entropy encoding unit (150) can use an encoding method such as exponential Golomb, Context-Adaptive Variable Length Coding (CAVLC), or Context-Adaptive Binary Arithmetic Coding (CABAC) for entropy encoding. For example, the entropy encoding unit (150) can perform entropy encoding using a Variable Length Coding / Code (VLC) table. In addition, the entropy encoding unit (150) may perform arithmetic encoding using the binarization method, probability model, and context model derived from the binarization method of the target symbol and the probability model of the target symbol / bin.

[0074] In this regard, when applying CABAC, the table probability update method can be changed to a simple formula-based table update method to reduce the size of the probability table stored in the decryption device. Furthermore, two different probability models can be used to obtain more accurate symbol probability values.

[0075] The entropy encoding unit (150) can change a two-dimensional block form coefficient into a one-dimensional vector form through a transform coefficient scanning method to encode a transform coefficient level (quantized level).

[0076] Coding parameters may include not only information (flags, indexes, etc.) encoded in an encoding device (100) and signaled to a decoding device (200), such as syntax elements, but also information derived during an encoding process or a decoding process, and may mean information necessary when encoding or decoding an image.

[0077] Here, signaling a flag or index may mean that the encoder entropy encodes the flag or index and includes it in the bitstream, and that the decoder entropy decodes the flag or index from the bitstream.

[0078] The encoded current image can be used as a reference image for other images to be processed later. Accordingly, the encoding device (100) can reconstruct or decode the encoded current image again and store the reconstructed or decoded image as a reference image in the reference picture buffer (190).

[0079] The quantized level can be dequantized in the dequantization unit (160) and inversely transformed in the inverse transformation unit (170). The dequantized and / or inversely transformed coefficients can be combined with a prediction block through an adder (117), and a reconstructed block can be generated by combining the dequantized and / or inversely transformed coefficients and the prediction block. Here, the dequantized and / or inversely transformed coefficients refer to coefficients on which at least one of dequantization and inverse transformation has been performed, and may refer to a reconstructed residual block. The dequantization unit (160) and the inverse transformation unit (170) can be performed in the reverse process of the quantization unit (140) and the transformation unit (130).

[0080] The restoration block may pass through a filter unit (180). The filter unit (180) may apply a deblocking filter, a sample adaptive offset (SAO), an adaptive loop filter (ALF), a bilateral filter (BIF), a Luma Mapping with Chroma Scaling (LMCS), etc. as a filtering technique, in whole or in part, to the restoration sample, restoration block, or restoration image. The filter unit (180) may also be referred to as an in-loop filter. In this case, the in-loop filter is also used as a name excluding LMCS.

[0081] A deblocking filter can remove block distortion that occurs at the boundaries between blocks. Whether to apply a deblocking filter to the current block can be determined based on the samples contained in several columns or rows within the block. When applying a deblocking filter to a block, different filters can be applied depending on the required deblocking filtering strength.

[0082] Sample adaptive offset can be used to compensate for encoding errors by adding an appropriate offset value to sample values. Sample adaptive offset can compensate for the offset from the original image on a sample-by-sample basis for deblocked images. This can be done by dividing the samples contained in the image into a fixed number of regions, determining the regions to be offset, and applying the offset to those regions. Alternatively, the offset can be applied by considering the edge information of each sample.

[0083] Bilateral filter (BIF) can also compensate for the offset from the original image on a sample-by-sample basis for the deblocked image.

[0084] An adaptive loop filter can perform filtering based on a comparison between a reconstructed image and the original image. By dividing the samples contained in the image into predetermined groups and determining the filter to be applied to each group, filtering can be performed differentially for each group. Information regarding whether to apply an adaptive loop filter can be signaled for each coding unit (CU), and the shape and filter coefficients of the adaptive loop filter applied to each block can vary.

[0085] In LMCS (Luma Mapping with Chroma Scaling), luma mapping (LM) refers to remapping luminance values ​​through a piece-wise linear model, and chroma scaling (CS) refers to a technique that scales the residual values ​​of chrominance components according to the average luminance value of the prediction signal. In particular, LMCS can be utilized as an HDR correction technique that reflects the characteristics of HDR (High Dynamic Range) images.

[0086] The restored block or restored image that has passed through the filter unit (180) may be stored in the reference picture buffer (190). The restored block that has passed through the filter unit (180) may be a part of the reference image. In other words, the reference image may be a restored image composed of restored blocks that have passed through the filter unit (180). The stored reference image may be used for inter-screen prediction or motion compensation thereafter.

[0087] FIG. 2 is a block diagram showing a configuration according to one embodiment of a decryption device to which the present disclosure is applied.

[0088] The decoding device (200) may be a decoder, a video decoding device, or an image decoding device.

[0089] Referring to FIG. 2, the decoding device (200) may include an entropy decoding unit (210), an inverse quantization unit (220), an inverse transformation unit (230), an intra prediction unit (240), a motion compensation unit (250), an adder (201), a switch (203), a filter unit (260), and a reference picture buffer (270).

[0090] The decoding device (200) can receive a bitstream output from the encoding device (100). The decoding device (200) can receive a bitstream stored in a computer-readable recording medium, or a bitstream streamed through a wired / wireless transmission medium. The decoding device (200) can perform decoding on the bitstream in intra mode or inter mode. In addition, the decoding device (200) can generate a restored image or a decoded image through decoding, and can output the restored image or the decoded image.

[0091] If the prediction mode used for decryption is intra mode, the switch (203) can be switched to intra. If the prediction mode used for decryption is inter mode, the switch (203) can be switched to inter.

[0092] The decoding device (200) can decode the input bitstream to obtain a reconstructed residual block and generate a prediction block. Once the reconstructed residual block and the prediction block are obtained, the decoding device (200) can generate a reconstructed block to be decoded by adding the reconstructed residual block and the prediction block. The block to be decoded may be referred to as a current block.

[0093] The entropy decoding unit (210) can generate symbols by performing entropy decoding according to a probability distribution for the bitstream. The generated symbols may include symbols in the form of quantized levels. Here, the entropy decoding method may be the reverse process of the entropy encoding method described above.

[0094] The entropy decoding unit (210) can change a one-dimensional vector-shaped coefficient into a two-dimensional block-shaped coefficient through a transform coefficient scanning method to decode a transform coefficient level (quantized level).

[0095] The quantized level can be inversely quantized in the inverse quantization unit (220) and inversely transformed in the inverse transformation unit (230). The quantized level can be generated as a restored residual block as a result of performing inverse quantization and / or inverse transformation. At this time, the inverse quantization unit (220) can apply a quantization matrix to the quantized level. The inverse quantization unit (220) and inverse transformation unit (230) applied to the decoding device can apply the same technology as the inverse quantization unit (160) and inverse transformation unit (170) applied to the encoding device described above.

[0096] When intra mode is used, the intra prediction unit (240) can generate a predicted block by performing spatial prediction on the current block using sample values ​​of already decoded blocks surrounding the block to be decoded. The intra prediction unit (240) applied to the decoding device can apply the same technology as the intra prediction unit (120) applied to the encoding device described above.

[0097] When the inter mode is used, the motion compensation unit (250) can generate a prediction block by performing motion compensation using a motion vector and a reference image stored in the reference picture buffer (270) on the current block. The motion compensation unit (250) can generate a prediction block by applying an interpolation filter to a portion of the reference image when the value of the motion vector does not have an integer value. In order to perform motion compensation, it is possible to determine whether the motion compensation method of the prediction unit included in the corresponding encoding unit is skip mode, merge mode, AMVP mode, or current picture reference mode based on the encoding unit, and motion compensation can be performed according to each mode. The motion compensation unit (250) applied to the decoding device can apply the same technology as the motion compensation unit (122) applied to the encoding device described above.

[0098] The adder (201) can add the restored residual block and the predicted block to generate a restored block. The filter unit (260) can apply at least one of an Inverse-LMCS, a deblocking filter, a sample adaptive offset, and an adaptive loop filter to the restored block or restored image. The filter unit (260) applied to the decoding device can apply the same filtering technology as that applied to the filter unit (180) applied to the encoding device described above.

[0099] The filter unit (260) can output a restored image. The restored block or restored image can be stored in the reference picture buffer (270) and used for inter prediction. The restored block that has passed through the filter unit (260) can be a part of the reference image. In other words, the reference image can be a restored image composed of restored blocks that have passed through the filter unit (260). The stored reference image can be used for inter-screen prediction or motion compensation thereafter.

[0100] FIG. 3 is a diagram schematically illustrating a video coding system to which the present disclosure can be applied.

[0101] A video coding system according to one embodiment may include an encoding device (10) and a decoding device (20). The encoding device (10) may transmit encoded video and / or image information or data to the decoding device (20) in the form of a file or streaming through a digital storage medium or a network.

[0102] An encoding device (10) according to one embodiment may include a video source generation unit (11), an encoding unit (12), and a transmission unit (13). A decoding device (20) according to one embodiment may include a reception unit (21), a decoding unit (22), and a rendering unit (23). The encoding unit (12) may be referred to as a video / image encoding unit, and the decoding unit (22) may be referred to as a video / image decoding unit. The transmission unit (13) may be included in the encoding unit (12). The reception unit (21) may be included in the decoding unit (22). The rendering unit (23) may include a display unit, and the display unit may be configured as a separate device or an external component.

[0103] The video source generation unit (11) can obtain video / images through a process of capturing, synthesizing, or generating video / images. The video source generation unit (11) can include a video / image capture device and / or a video / image generation device. The video / image capture device can include, for example, one or more cameras, a video / image archive including previously captured video / images, etc. The video / image generation device can include, for example, a computer, a tablet, a smartphone, etc., and can (electronically) generate video / images. For example, a virtual video / image can be generated through a computer, etc., in which case the video / image capture process can be replaced with a process of generating related data.

[0104] The encoding unit (12) can encode the input video / image. The encoding unit (12) can perform a series of procedures such as prediction, transformation, and quantization for compression and encoding efficiency. The encoding unit (12) can output encoded data (encoded video / image information) in the form of a bitstream. The detailed configuration of the encoding unit (12) can also be configured in the same manner as the encoding device (100) of FIG. 1 described above.

[0105] The transmission unit (13) can transmit encoded video / image information or data output in the form of a bitstream to the reception unit (21) of the decoding device (20) via a digital storage medium or a network in the form of a file or streaming. The digital storage medium can include various storage media such as USB, SD, CD, DVD, Blu-ray, HDD, SSD, etc. The transmission unit (13) can include an element for generating a media file through a predetermined file format and can include an element for transmission via a broadcasting / communication network. The reception unit (21) can extract / receive the bitstream from the storage medium or network and transmit it to the decoding unit (22).

[0106] The decoding unit (22) can decode video / image by performing a series of procedures such as inverse quantization, inverse transformation, and prediction corresponding to the operation of the encoding unit (12). The detailed configuration of the decoding unit (22) can also be configured in the same manner as the decoding device (200) of FIG. 2 described above.

[0107] The rendering unit (23) can render the decrypted video / image. The rendered video / image can be displayed through the display unit.

[0108]

[0109] Hereinafter, with reference to FIGS. 4 to 10, a deep fake detection method according to an embodiment of the present disclosure will be described.

[0110] In this specification, "deepfake" refers to a technology that uses deep learning technology to create manipulated images by superimposing other images on top of an image. A "deepfake video" may refer to a video manipulated using deepfake technology. Furthermore, "deepfake detection" refers to detecting whether deepfake technology has been applied to a video, and may be synonymous with "detecting whether a video is deepfake," "detecting deepfake videos," "determining whether a video is deepfake," or "determining whether a video is deepfake."

[0111] Additionally, in this specification, a syntactic element may be referred to as syntactic information.

[0112] As deepfakes are increasingly used across diverse fields, concerns about the technology are growing. While video manipulation in the past required significant time and expense, the emergence of deepfakes has made it easy for anyone to manipulate videos, and this technology can be exploited. Specifically, videos manipulated with deepfakes can cause social unrest or create false information, leading to social instability. Furthermore, creating fake videos using someone's face can cause significant emotional distress to the victim and can even lead to financial gain. Therefore, detecting deepfake videos and preventing such abuse is crucial.

[0113]

[0114] FIG. 4 is a diagram illustrating a deepfake detection method using syntactic elements of a bitstream according to an embodiment of the present disclosure. The deepfake detection method of FIG. 4 may be performed by a deepfake detection device. Here, the deepfake detection device may be a video decoding device or a part of a video decoding device. Additionally, the deepfake detection device may be a separate device connected to the video decoding device.

[0115] Referring to FIG. 4, the deepfake detection device can receive a bitstream of a video from a video encoding device or a streaming server (S410).

[0116] In addition, the deepfake detection device can obtain a syntax element used to decode an image from the received bitstream (S420) and calculate an unreliability measurement value based on the obtained syntax element (S430).

[0117] And, the deepfake detection device can detect deepfake based on the calculated unreliability measurement value (S440).

[0118] Here, the syntax elements for deepfake detection can be any one of quantization parameters, transformed coefficients, bit rate, intra prediction mode, motion vector, block vector, reference frame index, and boundary strength of a deblocking filter.

[0119]

[0120] When the syntax element for deepfake detection according to one embodiment of the present disclosure is a quantization parameter, the quantization parameters for a predetermined region within the current frame can be used for deepfake detection. Here, the predetermined region can be a region of interest (ROI) or the entire frame.

[0121] Meanwhile, if the syntactic element for deepfake detection is a quantization parameter, the unreliability measure can be calculated based on the difference between the quantization parameters within a predetermined region. Specifically, the unreliability measure can be calculated in proportion to the difference between the quantization parameters within the predetermined region, and if the unreliability measure is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0122] For example, when comparing the average value of the quantization parameter values ​​for N blocks constituting a predetermined area with the average value of the quantization parameter values ​​for some M blocks included in the predetermined area, an unreliability measure can be calculated based on the difference between these two values. Then, if the calculated unreliability measure is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0123] Meanwhile, when calculating the unreliability measure, in addition to the mean, any of the following values ​​can be used: median, mode, minimum, or maximum. Here, M can be an integer less than N, and when M is 1, the quantization parameter value for the corresponding block can be used for comparison.

[0124] In another embodiment of the present disclosure, when a syntax element for deepfake detection is a quantization parameter, the quantization parameters of frames in a predetermined section from a received video bitstream can be used for deepfake detection. Here, the predetermined section can be a frame interval of interest (FOI) or an entire video frame.

[0125] Meanwhile, when the syntactic element for deepfake detection is a quantization parameter, the unreliability measure can be calculated based on the difference in the quantization parameters of frames within a predetermined interval. Specifically, the unreliability measure can be calculated in proportion to the difference in the quantization parameters within the predetermined interval, and if the unreliability measure is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0126] For example, the average value of the quantization parameter values ​​for N frames constituting a predetermined section can be compared with the average value of the quantization parameter values ​​for some M frames included in the predetermined section, and an unreliability measurement value can be calculated based on the difference between these two values. Then, if the calculated unreliability measurement value is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0127] Meanwhile, when calculating the unreliability measure, any of the following values ​​can be used: median, mode, minimum, or maximum, in addition to the mean. Here, M can be an integer less than N.

[0128]

[0129] When a syntax element for deepfake detection according to an embodiment of the present disclosure is a bit rate, bit rates of frames in a predetermined section from a received video bitstream can be used for deepfake detection. Here, the predetermined section can be a frame section of interest or an entire video frame.

[0130] Meanwhile, when the syntactic element for deepfake detection is bitrate, the unreliability measurement value can be calculated based on the difference in bitrates of frames within a predetermined interval. Specifically, the unreliability measurement value can be calculated in proportion to the difference in bitrates within the predetermined interval, and if the unreliability measurement value is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0131] For example, the average bit rates of N frames constituting a predetermined section can be compared with the average bit rates of some M frames included in the predetermined section, and an unreliability measure can be calculated based on the difference between these two values. If the calculated unreliability measure is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0132] Meanwhile, when calculating the unreliability measure, any of the following values ​​can be used: median, mode, minimum, or maximum, in addition to the mean. Here, M can be an integer less than N.

[0133]

[0134] When the syntax element for deepfake detection according to one embodiment of the present disclosure is a motion vector, motion vectors for a predetermined region within the current frame can be used for deepfake detection. Here, the predetermined region can be a region of interest or the entire frame.

[0135] Meanwhile, when the syntactic element for deepfake detection is a motion vector, the unreliability measure can be calculated based on the difference in quantization parameters within a predetermined region. Specifically, the unreliability measure can be calculated in proportion to the difference in motion vectors within a predetermined region, and if the unreliability measure is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0136] For example, when comparing the average of motion vector values ​​for N blocks constituting a predetermined area with the average of motion vector values ​​for some M blocks included in the predetermined area, an unreliability measure can be calculated based on the difference between these two values. Then, if the calculated unreliability measure is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0137] Meanwhile, when calculating the unreliability measure, in addition to the average, one of the median, mode, minimum, and maximum values ​​can be used. Here, M can be an integer smaller than N, and when M is 1, the motion vector value for the corresponding block can be used for comparison. At this time, the x-direction motion vector and the y-direction motion vector can be compared respectively.

[0138] In another embodiment of the present disclosure, when the syntax element for deepfake detection is a motion vector, the unreliability measure may be calculated based on a motion vector obtained in units of a minimum block or in units of a block of a predetermined size.

[0139] Specifically, the unreliability measure can be calculated based on the difference between the motion vectors within a predetermined region and the motion vectors of co-located regions (i.e., temporally neighboring regions) of this region. Then, if the calculated unreliability measure is greater than or equal to a predetermined value, the current video can be determined as a deepfake video. For example, when comparing the average value of the motion vectors for blocks constituting a predetermined region with the average value of the motion vectors of co-located blocks of these blocks, the unreliability measure can be calculated based on the difference between these two values. Then, if the calculated unreliability measure is greater than or equal to a predetermined value, the current video can be determined as a deepfake video. Meanwhile, when calculating the unreliability measure, in addition to the average value, any one of the median, mode, minimum, and maximum values ​​can be used. At this time, the x-direction motion vector and the y-direction motion vector can be compared respectively.

[0140] Meanwhile, when calculating the unreliability measurement value, the motion vector may be a value processed by any one of mathematical formulas 1 to 4.

[0141] And, the above motion vector embodiment can be equally applied even when the syntax element for deepfake detection is a block vector.

[0142]

[0143] Mathematical formula 1

[0144] min(mv x , mv y )

[0145]

[0146] Mathematical formula 2

[0147] max(mv x , mv y )

[0148]

[0149] Mathematical formula 3

[0150] mv x 2 _ mv y 2

[0151]

[0152] Mathematical formula 4

[0153] |mv x | + |mv y |

[0154]

[0155] Meanwhile, when calculating unreliability measurements using motion vectors, motion vectors smaller than a certain block size can be excluded from the measurement (e.g., 4x4 blocks). This is because smaller blocks may not follow the overall motion flow of a specific object and may exhibit significant fluctuations in difference values, similar to noise.

[0156]

[0157] In one embodiment of the present disclosure, when the syntax element for deepfake detection is a transformation coefficient, the transformation coefficients for a predetermined region within the current frame can be used for deepfake detection. Here, the predetermined region can be a region of interest or the entire frame, and the transformation coefficients can be obtained in transformation units.

[0158] Meanwhile, when the syntactic element for deepfake detection is a transformation coefficient, the unreliability measure can be calculated based on the difference between the transformation coefficients within a predetermined region. Specifically, the unreliability measure can be calculated in proportion to the difference between the transformation coefficients within a predetermined region, and if the unreliability measure is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0159] For example, when comparing the average value of the transformation coefficient values ​​for N blocks constituting a predetermined area with the average value of the transformation coefficient values ​​for some M blocks included in the predetermined area, an unreliability measure can be calculated based on the difference between these two values. Then, if the calculated unreliability measure is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0160] Meanwhile, when calculating the unreliability measure, in addition to the mean, any of the median, mode, minimum, or maximum values ​​can be used. Here, M can be an integer less than N, and when M is 1, the transformation coefficient value for the corresponding transformation block can be used for comparison.

[0161] In another embodiment of the present disclosure, when the syntax element for deepfake detection is a transform coefficient, the unreliability measure can be calculated based on the transform coefficient obtained in units of transform blocks for each frame.

[0162] Specifically, the unreliability measure can be calculated based on the difference between the average value of the transform coefficients at a specific location within the transform blocks constituting a predetermined region and the average value of the transform coefficients at a specific location within the co-located transform blocks of these transform blocks (i.e., temporally neighboring transform blocks). Then, if the calculated unreliability measure is greater than a predetermined value, the current video can be determined to be a deepfake video. Meanwhile, when calculating the unreliability measure, in addition to the average value, any one of the median, mode, minimum, and maximum values ​​can be used. At this time, the x-direction motion vector and the y-direction motion vector can be compared respectively.

[0163] Here, when measuring unreliability based on the transformation coefficients of the transformation blocks, intra-blocks and inter-blocks can be compared separately. Because the absolute value of the change coefficients of intra-blocks is relatively larger than that of inter-blocks, the two prediction types can be separated and the differences in their transformation coefficients compared.

[0164] Meanwhile, a specific location of a transformation block can be determined based on the scan order of the transformation coefficients.

[0165] Figure 5 shows an example of a transform coefficient at a specific location within a transform block when the size of the transform block is 4x4 and the scan order of the transform coefficients is a diagonal scan order. Referring to Figure 5, when six high-frequency transform coefficients are used for deepfake detection, six transform coefficients (C10 to C15) can be used in the reverse order of the diagonal scan order, starting from C15 located at the lower right of the transform block.

[0166] Figure 6 shows an example of a transform coefficient at a specific location within a transform block when the size of the transform block is 4x4 and the scan order of the transform coefficients is a horizontal scan order. Referring to Figure 6, when six high-frequency transform coefficients are used for deepfake detection, six transform coefficients (C10 to C15) can be used in the reverse order of the horizontal scan order, starting from C15 located at the lower right of the transform block.

[0167]

[0168] When a syntax element for deepfake detection according to an embodiment of the present disclosure is a reference frame index, reference frame indices for a predetermined region within the current frame can be used for deepfake detection. Here, the predetermined region can be a region of interest or the entire frame, and the reference frame indices can be obtained in units of minimum blocks or blocks of a predetermined size.

[0169] Meanwhile, if the syntactic element for deepfake detection is a reference frame index, the unreliability measure can be calculated based on the difference between reference frame indices within a predetermined region. Specifically, the unreliability measure can be calculated in proportion to the difference between reference frame indices within a predetermined region, and if the unreliability measure is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0170] For example, when comparing the average value of the reference frame indices for N blocks constituting a predetermined area with the average value of the reference frame indices for some M blocks included in the predetermined area, an unreliability measure can be calculated based on the difference between these two values. Then, if the calculated unreliability measure is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0171] Meanwhile, when calculating the unreliability measure, in addition to the mean, any of the following values ​​can be used: median, mode, minimum, or maximum. Here, M can be an integer less than N, and if M is 1, the reference frame index for the corresponding block can be used for comparison.

[0172] At this time, considering the syntactic elements indicating the reference direction, the reference frame indices can be compared only when they are in the same reference direction. In the case of bidirectional prediction, the reference frame indices in the L0 direction and the reference frame indices in the L1 direction can be compared respectively. Here, L0 means a list of reference pictures preceding the current picture in playback order, and L1 means a list of reference pictures following the current picture in playback order.

[0173]

[0174] In an embodiment of the present disclosure, when a syntax element for deepfake detection is a boundary strength of a deblocking filter, boundary strengths of the deblocking filter for a predetermined region within the current frame can be used for deepfake detection. Here, the predetermined region can be a region of interest or the entire frame, and the boundary strength of the deblocking filter can be obtained in units of minimum blocks or blocks of a predetermined size.

[0175] Meanwhile, if the syntactic element for deepfake detection is the boundary strength of the deblocking filter, the unreliability measurement value can be calculated based on the difference in the boundary strengths of the deblocking filter within a predetermined area. Specifically, the unreliability measurement value can be calculated in proportion to the difference in the boundary strengths of the deblocking filter within a predetermined area, and if the unreliability measurement value is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0176] For example, when comparing the average value of the boundary intensities of the deblocking filter for N blocks constituting a predetermined area with the average value of the boundary intensities of the deblocking filter for some M blocks included in the predetermined area, an unreliability measure can be calculated based on the difference between these two values. Then, if the calculated unreliability measure is greater than a predetermined value, the current video can be determined to be a deepfake video.

[0177] Meanwhile, when calculating the unreliability measure, in addition to the mean, any of the following values ​​can be used: median, mode, minimum, or maximum. Here, M can be an integer less than N, and when M is 1, the boundary strength of the deblocking filter for the corresponding block can be used for comparison.

[0178] In another embodiment of the present disclosure, when the syntax element for deepfake detection is the boundary strength of a deblocking filter, the unreliability measurement value can be calculated based on the boundary strength of the deblocking filter obtained in units of minimum blocks or blocks of a predetermined size.

[0179] Specifically, the unreliability measure can be calculated based on the difference between the edge intensities of the deblocking filter within a predetermined region and the edge intensities of the deblocking filter in co-located regions (i.e., temporally neighboring regions) of this region. Then, if the calculated unreliability measure is greater than or equal to a predetermined value, the current video can be determined as a deepfake video. For example, when comparing the average value of the edge intensities of the deblocking filter for blocks constituting the predetermined region with the average value of the edge intensities of the deblocking filters of blocks co-located with these blocks, the unreliability measure can be calculated based on the difference between these two values. Then, if the calculated unreliability measure is greater than or equal to a predetermined value, the current video can be determined as a deepfake video. Meanwhile, when calculating the unreliability measure, in addition to the average value, any one of the median, the mode, the minimum, and the maximum values ​​can be used.

[0180]

[0181] A deepfake detection method according to one embodiment of the present disclosure may utilize multiple syntactic elements rather than one syntactic element.

[0182] FIG. 7 is a diagram illustrating a deepfake detection method using multiple syntax elements according to one embodiment of the present disclosure.

[0183] Referring to FIG. 7, the deepfake detection device can receive a bitstream of a video from a video encoding device or a streaming server (S710).

[0184] In addition, the deepfake detection device can obtain a first syntax element and a second syntax element used to decode an image from a received bitstream (S720), calculate a first unreliability measurement value based on the obtained first syntax element, and calculate a second unreliability measurement value based on the second syntax element (S730).

[0185] And, the deepfake detection device can detect deepfake based on the first unreliability measurement value and the second unreliability measurement value produced (S740).

[0186] In Fig. 7, an embodiment using the first syntax element and the second syntax element is described, but X syntax elements may be used. Here, X may be an integer greater than or equal to 2.

[0187] Meanwhile, when detecting deepfakes using multiple syntactic elements, deepfakes can be detected using multiple syntactic elements in a predetermined order.

[0188] For example, a deepfake detection device can calculate a first unreliability measurement value using a first syntax element, and detect whether a deepfake is present based on the calculated first unreliability measurement value. If the deepfake detection device determines that the video is a deepfake, the detection ends, and if the video is not determined to be a deepfake, the detection process continues. Next, the deepfake detection device can calculate a second unreliability measurement value using the acquired second syntax element, and detect whether a deepfake is present based on the calculated second unreliability measurement value. If the deepfake detection device determines that the video is a deepfake, the detection ends, and if the video is not determined to be a deepfake, the detection process continues. This process can continue up to a detection step using the Xth unreliability measurement value.

[0189]

[0190] A deepfake detection method according to one embodiment of the present disclosure may utilize an unreliability score calculated by weighting multiple unreliability measurements.

[0191] FIG. 8 is a diagram illustrating a deepfake detection method using an unreliability score according to an embodiment of the present disclosure.

[0192] Referring to FIG. 8, the deepfake detection device can receive a bitstream of a video from a video encoding device or a streaming server (S810).

[0193] In addition, the deepfake detection device can obtain a first syntax element and a second syntax element used to decode an image from a received bitstream (S820), calculate a first unreliability measurement value based on the obtained first syntax element, and calculate a second unreliability measurement value based on the second syntax element (S830).

[0194] In addition, the deepfake detection device can calculate a distrust score by weighting the calculated first distrust measurement value and the calculated second distrust measurement value (S840). Here, the weight of the weighted sum may be a weight preset for the deepfake detection device.

[0195] And, the deepfake detection device can detect deepfake based on the calculated unreliability score (S850).

[0196] In Fig. 8, an embodiment using the first syntax element and the second syntax element is described, but Y syntax elements may be used. Here, Y may be an integer greater than or equal to 2.

[0197] Preset weights (w1, w2…, w) as in mathematical formula 5 Y ) is applied to the unreliability measure to obtain the unreliability score (UR) of the image. Score ) can be produced.

[0198]

[0199] Mathematical Formula 5

[0200] UR Score = w1*x1+ w2*x2+ … + w Y *x Y

[0201]

[0202] A deepfake detection method according to one embodiment of the present disclosure may utilize an unreliability score calculated by normalizing a plurality of unreliability measurements.

[0203] FIG. 9 is a diagram illustrating a deepfake detection method based on a normalized unreliability measurement value according to one embodiment of the present disclosure.

[0204] Referring to FIG. 9, the deepfake detection device can receive a bitstream of a video from a video encoding device or a streaming server (S910).

[0205] And, the deepfake detection device can obtain a first syntax element and a second syntax element used to decode an image from a received bitstream (S920), calculate a first unreliability measurement value based on the obtained first syntax element, and calculate a second unreliability measurement value based on the second syntax element (S930).

[0206] In addition, the deepfake detection device can normalize the first unreliable measurement value and the second unreliable measurement value produced (S940).

[0207] In addition, the deepfake detection device can calculate a distrust score by weighting the normalized first distrust measurement value and the second distrust measurement value (S950). Here, the weight of the weighted sum may be a weight preset for the deepfake detection device.

[0208] And, the deepfake detection device can detect deepfake based on the calculated unreliability score (S960).

[0209] In Fig. 9, an embodiment using the first syntax element and the second syntax element is described, but Z syntax elements may be used. Here, Z may be an integer greater than or equal to 2.

[0210]

[0211] A deepfake detection method according to one embodiment of the present disclosure may utilize syntactic elements of a region of interest (ROI).

[0212] Here, the region of interest may be specified in advance, or a process for determining the region of interest may be additionally performed.

[0213] For example, when performing image decoding, a region in which a motion vector exceeds a predefined value or / and a region in which a transform coefficient exceeds a predefined value can be determined as a region of interest.

[0214] As another example, object detection can be performed on the decrypted image to determine the region of interest, after which deepfake detection methods can be performed. Object detection can target the facial region.

[0215]

[0216] A deepfake detection method according to one embodiment of the present disclosure can utilize spatial domain information of an image in addition to syntactic element information of a bitstream. Here, spatial domain information refers to information obtainable from the spatial domain of an image, and may include the degree of illumination variation, sharpness, contrast, flickering, and the like.

[0217] The deepfake detection device can simultaneously obtain a number of spatial domain information pieces from the decrypted image, while calculating the unreliability measurement value according to the previously described embodiment, and calculate the image's unreliability score by applying preset weights to the unreliability measurement values ​​and spatial domain information pieces. Even in this case, the unreliability measurement values ​​and spatial domain information pieces can be normalized and utilized prior to calculating the unreliability score.

[0218]

[0219] FIG. 10 is a flowchart illustrating a deepfake detection method according to an embodiment of the present disclosure. The deepfake detection method of FIG. 10 can be performed by a deepfake detection device.

[0220] Referring to FIG. 10, the deepfake detection device can receive a bitstream of a video from a video encoding device or a streaming server (S1010).

[0221] In addition, the deepfake detection device can obtain syntax information from the received bitstream (S1020). Here, the syntax information may be at least one of a quantization parameter, a transform coefficient, a bit rate, a motion vector, a reference picture index, or a boundary strength of a deblocking filter.

[0222] Meanwhile, the deepfake detection device can obtain syntactic information of the region of interest within the current picture.

[0223] According to one embodiment, a region of interest may be determined as a region in which a motion vector exceeding a predefined value exists.

[0224] According to another embodiment, the region of interest may be determined as a region in which a transformation coefficient greater than a predefined value exists.

[0225] In another embodiment, the region of interest may be determined based on object detection.

[0226] In addition, the deepfake detection device can calculate a distrust measurement value based on the acquired syntax information (S1030).

[0227] And, the deepfake detection device can determine whether the video is a deepfake based on the calculated unreliability measurement value (S1040).

[0228] Whether the video is a deepfake can be determined based on multiple unreliability measures calculated based on multiple pieces of syntactic information. In this case, whether the video is a deepfake can be determined based on a reliability score derived by applying a weight to each of the multiple unreliability measures. Here, the reliability score can be derived by normalizing each of the multiple unreliability measures and applying the weight.

[0229] Meanwhile, whether the video is a deepfake can be determined based on the unreliability measurement value and spatial domain information. In this case, whether the video is a deepfake can be determined based on an unreliability score derived by applying weights to each of the unreliability measurement value and the spatial domain information. Here, the spatial domain information may be at least one of illumination change information, sharpness information, contrast information, or blinking information.

[0230]

[0231] FIG. 11 is a block diagram illustrating a deepfake detection device according to an embodiment of the present disclosure.

[0232] Referring to FIG. 11, a deepfake detection device (1100) may include a receiving unit (1110) and a detection unit (1120).

[0233] The receiving unit (1110) can receive a bitstream of an image.

[0234] In addition, the detection unit (1120) can obtain syntax information from the bitstream, calculate an unreliability measurement value based on the syntax information, and determine whether the image is a deepfake based on the unreliability measurement value. Here, the syntax information can be at least one of a quantization parameter, a transform coefficient, a bit rate, a motion vector, a reference picture index, or a deblocking filter boundary strength.

[0235] The deepfake detection device (1100) may be a part included in the decryption device (200) of FIG. 2.

[0236] Alternatively, the deepfake detection device (1100) may be included in the decryption device (20) of FIG. 3. Specifically, the receiving unit (21) of FIG. 3 has the same configuration as the receiving unit (1110) of the deepfake detection device (1100), and the detection unit (1120) of the deepfake detection device (1100) may be configured to be separately connected to the decryption unit (22) of FIG. 3.

[0237]

[0238] FIG. 12 is a drawing exemplifying a content streaming system to which an embodiment according to the present disclosure can be applied.

[0239] As illustrated in FIG. 12, a content streaming system to which an embodiment of the present disclosure is applied may largely include an encoding server, a streaming server, a web server, a media storage, a user device, and a multimedia input device.

[0240] The encoding server compresses content input from multimedia input devices such as smartphones, cameras, and CCTVs into digital data, generates a bitstream, and transmits it to the streaming server. Alternatively, if multimedia input devices such as smartphones, cameras, and CCTVs directly generate bitstreams, the encoding server may be omitted.

[0241] The above bitstream can be generated by a video encoding method and / or a video encoding device, and the streaming server can temporarily store the bitstream during the process of transmitting or receiving the bitstream.

[0242] The streaming server transmits multimedia data to a user device based on a user request via a web server, and the web server can act as an intermediary to inform the user of available services. When a user requests a desired service from the web server, the web server transmits the request to the streaming server, and the streaming server can transmit multimedia data to the user. At this time, the content streaming system may include a separate control server, and in this case, the control server may control commands / responses between each device within the content streaming system.

[0243] The streaming server can receive content from a media repository and / or encoding server. For example, when receiving content from the encoding server, the content can be received in real time. In this case, to provide a smooth streaming service, the streaming server can store the bitstream for a certain period of time.

[0244] Examples of the user devices may include mobile phones, smart phones, laptop computers, digital broadcasting terminals, personal digital assistants (PDAs), portable multimedia players (PMPs), navigation devices, slate PCs, tablet PCs, ultrabooks, wearable devices (e.g., smartwatches, smart glasses, HMDs), digital TVs, desktop computers, digital signage, etc.

[0245] Each server within the above content streaming system can be operated as a distributed server, in which case data received from each server can be processed in a distributed manner.

[0246]

[0247] It is possible to detect whether a video is a deepfake using at least one or a combination of at least one of the above embodiments.

[0248] The above embodiments can be performed for each of the luminance and chrominance signals. Alternatively, the above embodiments can be performed identically for the luminance and chrominance signals.

[0249] In the above embodiments, the methods are described based on a flowchart as a series of steps or units. However, the present disclosure is not limited to the order of the steps, and some steps may occur in a different order or simultaneously with other steps described above. Furthermore, those skilled in the art will understand that the steps depicted in the flowchart are not exclusive, and that other steps may be included, or one or more steps in the flowchart may be deleted without affecting the scope of the present disclosure.

[0250] The above embodiments may be implemented in the form of program commands that can be executed by various computer components and recorded on a computer-readable recording medium. The computer-readable recording medium may include program commands, data files, data structures, etc., either singly or in combination. The program commands recorded on the computer-readable recording medium may be those specifically designed and configured for the present disclosure, or may be known and usable by those skilled in the art of computer software.

[0251] The bitstream generated by the encoding method according to the above embodiment can be stored in a non-transitory computer-readable recording medium. In addition, the bitstream stored in the non-transitory computer-readable recording medium can be decoded by the decoding method according to the above embodiment.

[0252] Here, examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specifically configured to store and execute program instructions such as ROMs, RAMs, and flash memories. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware devices may be configured to operate as one or more software modules to perform processing according to the present disclosure, and vice versa.

[0253] Although the present disclosure has been described above with specific details such as specific components and limited embodiments and drawings, these are provided only to help a more general understanding of the present disclosure, and the present disclosure is not limited to the above embodiments, and a person having ordinary knowledge in the technical field to which the present disclosure belongs can make various modifications and variations from this description.

[0254] Therefore, the spirit of the present disclosure should not be limited to the embodiments described above, and all modifications that are equivalent or equivalent to the following claims as well as the claims are considered to fall within the scope of the spirit of the present disclosure.

[0255]

[0256] The present invention can be used in a device for encoding / decoding an image, a deepfake detection device, and a recording medium storing a bitstream.

Claims

1. Step of receiving a bitstream of a video; A step of obtaining syntax information from the above bitstream; A step of calculating an unreliability measurement value based on the above syntax information; and A step of determining whether the video is a deepfake based on the above unreliability measurement value is included, A deepfake detection method, wherein the above syntax information is at least one of a quantization parameter, a transform coefficient, a bit rate, a motion vector, a reference picture index, or a boundary strength of a deblocking filter.

2. In paragraph 1, A deepfake detection method, characterized in that the step of obtaining the above syntax information obtains syntax information of a region of interest within the current picture.

3. In paragraph 2, A deepfake detection method, characterized in that the above region of interest is determined as a region in which a motion vector exceeding a predefined value exists.

4. In paragraph 2, A deepfake detection method, characterized in that the above region of interest is determined as a region in which a transformation coefficient greater than a predefined value exists.

5. In paragraph 2, A deepfake detection method, characterized in that the above region of interest is determined based on object detection.

6. In paragraph 1, A deepfake detection method characterized in that whether the above video is a deepfake is determined based on multiple unreliability measurement values ​​calculated based on multiple pieces of syntax information.

7. In paragraph 6, A deepfake detection method characterized in that whether the above video is a deepfake is determined based on a distrust score derived by applying a weight to each of a plurality of distrust measurement values.

8. In paragraph 7, A deepfake detection method, characterized in that the above unreliability score is derived by normalizing each of the plurality of unreliability measurements and applying the weight.

9. In paragraph 1, A deepfake detection method, characterized in that whether the above video is a deepfake is determined based on the unreliability measurement value and spatial domain information.

10. In paragraph 9, A deepfake detection method, wherein the spatial domain information is at least one of illumination change information, sharpness information, contrast information, or blinking information.

11. In paragraph 9, A deepfake detection method, characterized in that whether the above video is a deepfake is determined based on an unreliability score derived by applying a weight to each of the unreliability measurement value and the spatial domain information.

12. A receiving unit that receives a bitstream of an image; and Obtaining syntax information from the above bitstream, Based on the above syntax information, a non-reliability measurement value is calculated, A detection unit that determines whether the video is a deepfake based on the above unreliability measurement value is included, A deepfake detection device, characterized in that the above syntax information is at least one of a quantization parameter, a transform coefficient, a bit rate, a motion vector, a reference picture index, or a deblocking filter boundary strength.

Images