KR20230026944A - Video Coding Method And Apparatus Using Improved Inloop Filter - Google Patents

Abstract

A video coding method and apparatus using an improved in-loop filter are disclosed. In the present embodiment, a video coding method and apparatus are provided. In order to improve video encoding efficiency and video quality, a residual frame is generated from a restored frame using a deep learning model. The generated residual frame is applied to a linear model to approximate an original residual frame. Therefore, it is possible to improve the performance of the in-loop filter.

Description

Video Coding Method And Apparatus Using Improved Inloop Filter

The present disclosure relates to a video coding method and apparatus using an improved in-loop filter.

The contents described below merely provide background information related to the present embodiment and do not constitute prior art.

Since video data has a large amount of data compared to audio data or still image data, it requires a lot of hardware resources including memory to store or transmit itself without processing for compression.

Therefore, when video data is stored or transmitted, an encoder is used to compress and store or transmit the video data, and a decoder receives, decompresses, and reproduces the compressed video data. Examples of such video compression technologies include H.264/AVC, High Efficiency Video Coding (HEVC), and Versatile Video Coding (VVC), which has improved coding efficiency by about 30% or more compared to HEVC.

However, since the size, resolution, and frame rate of video are gradually increasing, and the amount of data to be encoded accordingly increases, a new compression technology with higher encoding efficiency and higher picture quality improvement effect than existing compression technologies is required.

A general picture quality improvement algorithm based on deep learning technology seeks to improve picture quality and reduce the amount of computation by using a residual signal along a skip path. In addition, deep learning-based image processing technology is applied to existing encoding element technology. Encoding efficiency can be improved by applying deep learning-based image processing technology to compression technologies such as inter prediction, intra prediction, in-loop filter, and transformation among existing encoding technologies. Representative application examples include inter prediction based on a virtual reference frame generated based on a deep learning model, in-loop filter based on a restored frame, and the like. Therefore, in image encoding/decoding, application of a deep learning-based image processing technology using a residual signal needs to be considered in order to improve encoding efficiency and image quality.

In the present disclosure, in order to improve video encoding efficiency and video quality, a residual frame is generated from a restored frame using a deep learning model, and the generated residual frame is applied to a linear model. It is an object of the present invention to provide a video coding method and apparatus for improving the performance of an in-loop filter by approximating an original residual frame.

According to an embodiment of the present disclosure, in a method for improving the quality of a reconstructed frame performed by a computing device, the reconstructed frame is acquired and the reconstructed frame is input to a deep learning-based improvement model. generating an output approximating an original frame, wherein the restored frame is a frame obtained by reconstructing the original frame, and is previously restored by the computing device; generating a first residual frame by subtracting the restored frame from the output of the improved model; generating a second residual frame by inputting the first residual frame into a linear model, wherein the linear model includes parameters indicating a linear relationship between the first residual frame and the second residual frame; and generating an improved reconstruction frame by adding the second residual frame and the reconstruction frame.

According to another embodiment of the present disclosure, in an image quality improving apparatus included in an image decoding apparatus, a deep learning-based improvement model is included, a reconstructed frame is obtained, and the reconstructed frame is converted into the improved model. a first enhancement unit for generating an output approximating an original frame by inputting an input to an original frame, wherein the restored frame is a frame obtained by restoring the original frame and is previously restored by the video decoding apparatus; a subtractor configured to subtract the reconstructed frame from the output of the improved model to generate a first residual frame; A second improvement unit including a linear model and generating a second residual frame by inputting the first residual frame to the linear model, wherein the linear model is a linear relationship between the first residual frame and the second residual frame including parameters representing; and an adder generating an improved reconstruction frame by adding the second residual frame and the reconstruction frame.

According to another embodiment of the present disclosure, in a method for improving the quality of a reconstructed frame performed by a computing device, the reconstructed frame is obtained, and the reconstructed frame is converted to a deep learning-based improvement model. generating an output approximating an original frame by inputting the restored frame, wherein the restored frame is a frame obtained by restoring the original frame, and is previously restored by the computing device; generating a first residual frame by subtracting the restored frame from the output of the improved model; Obtaining encoding/decoding information, and generating an embedding vector by inputting the encoding/decoding information into a deep learning-based embedding function, wherein the encoding/decoding information includes a quantization parameter and bit rate distortion. At least one of a Lagrange constant used in an optimization process, a temporal layer in a group of pictures (GOP), and a type of frame; generating a complemented residual frame by matrix multiplication of the embedding vector and the first residual frame; generating a second residual frame by inputting the augmented residual frame into a linear model, wherein the linear model includes parameters indicating a linear relationship between the augmented residual frame and the second residual frame; and generating an improved reconstruction frame by adding the second residual frame and the reconstruction frame.

As described above, according to the present embodiment, by providing a video coding method and apparatus for generating a residual frame from a restored frame using a deep learning model and approximating the original residual frame by applying the generated residual frame to a linear model, , there is an effect that it becomes possible to improve video encoding efficiency and video quality according to the performance improvement of the in-loop filter.

1 is an exemplary block diagram of an image encoding apparatus capable of implementing the techniques of this disclosure.
2 is a diagram for explaining a method of dividing a block using a QTBTTT structure.
3A and 3B are diagrams illustrating a plurality of intra prediction modes including wide-angle intra prediction modes.
4 is an exemplary diagram of neighboring blocks of a current block.
5 is an exemplary block diagram of a video decoding apparatus capable of implementing the techniques of this disclosure.
6 is an exemplary diagram illustrating a picture quality improving apparatus according to an embodiment of the present disclosure.
7 is an exemplary diagram illustrating a picture quality improving apparatus according to another embodiment of the present disclosure.
8 is an exemplary diagram illustrating a picture quality improving apparatus according to another embodiment of the present disclosure.
9 is a flowchart illustrating a method for improving an image quality according to an embodiment of the present disclosure.
10 is a flowchart illustrating a method for improving an image quality according to an embodiment of the present disclosure.

DETAILED DESCRIPTION Some embodiments of the present disclosure are described in detail below with reference to exemplary drawings. In adding reference numerals to components of each drawing, it should be noted that the same components have the same numerals as much as possible even if they are displayed on different drawings. In addition, in describing the present embodiments, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the present embodiments, the detailed description will be omitted.

1 is an exemplary block diagram of an image encoding apparatus capable of implementing the techniques of this disclosure. Hereinafter, with reference to the illustration of FIG. 1, an image encoding device and sub-components of the device will be described.

The image encoding apparatus includes a picture division unit 110, a prediction unit 120, a subtractor 130, a transform unit 140, a quantization unit 145, a rearrangement unit 150, an entropy encoding unit 155, and an inverse quantization unit. 160, an inverse transform unit 165, an adder 170, a loop filter unit 180, and a memory 190.

Each component of the image encoding device may be implemented as hardware or software, or as a combination of hardware and software. Also, the function of each component may be implemented as software, and the microprocessor may be implemented to execute the software function corresponding to each component.

One image (video) is composed of one or more sequences including a plurality of pictures. Each picture is divided into a plurality of areas and encoding is performed for each area. For example, one picture is divided into one or more tiles or/and slices. Here, one or more tiles may be defined as a tile group. Each tile or/slice is divided into one or more Coding Tree Units (CTUs). And each CTU is divided into one or more CUs (Coding Units) by a tree structure. Information applied to each CU is coded as a CU syntax, and information commonly applied to CUs included in one CTU is coded as a CTU syntax. In addition, information commonly applied to all blocks in one slice is coded as syntax of a slice header, and information applied to all blocks constituting one or more pictures is a picture parameter set (PPS) or picture coded in the header. Furthermore, information commonly referred to by a plurality of pictures is coded into a Sequence Parameter Set (SPS). Also, information commonly referred to by one or more SPSs is coded into a video parameter set (VPS). Also, information commonly applied to one tile or tile group may be encoded as syntax of a tile or tile group header. Syntax included in the SPS, PPS, slice header, tile or tile group header may be referred to as high level syntax.

The picture divider 110 determines the size of a coding tree unit (CTU). Information on the size of the CTU (CTU size) is encoded as SPS or PPS syntax and transmitted to the video decoding apparatus.

The picture division unit 110 divides each picture constituting an image into a plurality of Coding Tree Units (CTUs) having a predetermined size, and then iteratively divides the CTUs using a tree structure. Divide (recursively). A leaf node in the tree structure becomes a coding unit (CU), which is a basic unit of encoding.

As a tree structure, a quad tree (QT) in which a parent node (or parent node) is divided into four subnodes (or child nodes) of the same size, or a binary tree (BinaryTree) in which a parent node is divided into two subnodes , BT), or a TernaryTree (TT) in which a parent node is split into three subnodes at a ratio of 1:2:1, or a structure in which two or more of these QT structures, BT structures, and TT structures are mixed. there is. For example, a QuadTree plus BinaryTree (QTBT) structure may be used, or a QuadTree plus BinaryTree TernaryTree (QTBTTT) structure may be used. Here, BTTT may be combined to be referred to as MTT (Multiple-Type Tree).

2 is a diagram for explaining a method of dividing a block using a QTBTTT structure.

As shown in FIG. 2, the CTU may first be divided into QT structures. Quadtree splitting can be repeated until the size of the splitting block reaches the minimum block size (MinQTSize) of leaf nodes allowed by QT. A first flag (QT_split_flag) indicating whether each node of the QT structure is split into four nodes of a lower layer is encoded by the entropy encoder 155 and signaled to the video decoding device. If the leaf node of QT is not larger than the maximum block size (MaxBTSize) of the root node allowed in BT, it may be further divided into either a BT structure or a TT structure. A plurality of division directions may exist in the BT structure and/or the TT structure. For example, there may be two directions in which blocks of the corresponding node are divided horizontally and vertically. As shown in FIG. 2, when MTT splitting starts, a second flag (mtt_split_flag) indicating whether nodes are split, and if split, a flag indicating additional split direction (vertical or horizontal) and/or split type (Binary or Ternary) is encoded by the entropy encoding unit 155 and signaled to the video decoding apparatus.

Alternatively, prior to coding the first flag (QT_split_flag) indicating whether each node is split into four nodes of a lower layer, a CU split flag (split_cu_flag) indicating whether the node is split is coded. It could be. When the value of the CU split flag (split_cu_flag) indicates that it is not split, the block of the corresponding node becomes a leaf node in the split tree structure and becomes a coding unit (CU), which is a basic unit of encoding. When the value of the CU split flag (split_cu_flag) indicates splitting, the video encoding apparatus starts encoding from the first flag in the above-described manner.

As another example of a tree structure, when QTBT is used, the block of the corresponding node is divided into two blocks of the same size horizontally (i.e., symmetric horizontal splitting) and the type that splits vertically (i.e., symmetric vertical splitting). Branches may exist. A split flag (split_flag) indicating whether each node of the BT structure is split into blocks of a lower layer and split type information indicating a split type are encoded by the entropy encoder 155 and transmitted to the video decoding device. Meanwhile, a type in which a block of a corresponding node is divided into two blocks having an asymmetric shape may additionally exist. 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 a form in which the block of the corresponding node is divided in a diagonal direction may be included.

A CU can have various sizes depending on the QTBT or QTBTTT split from the CTU. Hereinafter, a block corresponding to a CU to be encoded or decoded (ie, a leaf node of QTBTTT) is referred to as a 'current block'. Depending on the adoption of the QTBTTT division, the shape of the current block may be rectangular as well as square.

The prediction unit 120 predicts a current block and generates a prediction block. The prediction unit 120 includes an intra prediction unit 122 and an inter prediction unit 124 .

In general, each current block in a picture can be coded predictively. In general, prediction of a current block uses an intra-prediction technique (using data from a picture containing the current block) or an inter-prediction technique (using data from a picture coded before the picture containing the current block). can be performed Inter prediction includes both uni-prediction and bi-prediction.

The intra predictor 122 predicts pixels in the current block using pixels (reference pixels) located around the current block in the current picture including the current block. A plurality of intra prediction modes exist according to the prediction direction. For example, as shown in FIG. 3A, the plurality of intra prediction modes may include two non-directional modes including a planar mode and a DC mode and 65 directional modes. Depending on each prediction mode, the neighboring pixels to be used and the arithmetic expression are defined differently.

For efficient directional prediction of the rectangular current block, directional modes (numbers 67 to 80 and -1 to -14 intra prediction modes) indicated by dotted arrows in FIG. 3B may be additionally used. These may be referred to as “wide angle intra-prediction modes”. In FIG. 3B , arrows indicate corresponding reference samples used for prediction and do not indicate prediction directions. The prediction direction is opposite to the direction the arrow is pointing. Wide-angle intra prediction modes are modes that perform prediction in the opposite direction of a specific directional mode without additional bit transmission when the current block is rectangular. At this time, among the wide-angle intra prediction modes, some wide-angle intra prediction modes usable for the current block may be determined by the ratio of the width and height of the rectangular current block. For example, wide-angle intra prediction modes (67 to 80 intra prediction modes) having an angle smaller than 45 degrees are usable when the current block has a rectangular shape with a height smaller than a width, and a wide angle having an angle greater than -135 degrees. Intra prediction modes (-1 to -14 intra prediction modes) are available when the current block has a rectangular shape where the width is greater than the height.

The intra prediction unit 122 may determine an intra prediction mode to be used for encoding the current block. In some examples, the intra prediction unit 122 may encode the current block using several intra prediction modes and select an appropriate intra prediction mode to be used from the tested modes. For example, the intra predictor 122 calculates rate-distortion values using rate-distortion analysis for several tested intra-prediction modes, and has the best rate-distortion characteristics among the tested modes. Intra prediction mode can also be selected.

The intra prediction unit 122 selects one intra prediction mode from among a plurality of intra prediction modes, and predicts a current block using neighboring pixels (reference pixels) determined according to the selected intra prediction mode and an arithmetic expression. Information on the selected intra prediction mode is encoded by the entropy encoder 155 and transmitted to the video decoding apparatus.

The inter prediction unit 124 generates a prediction block for a current block using a motion compensation process. The inter-prediction unit 124 searches for a block most similar to the current block in the encoded and decoded reference picture prior to the current picture, and generates a prediction block for the current block using the searched block. Then, a motion vector (MV) corresponding to displacement between the current block in the current picture and the prediction block in the reference picture is generated. In general, motion estimation is performed on a luma component, and a motion vector calculated based on the luma component is used for both the luma component and the chroma component. Motion information including reference picture information and motion vector information used to predict the current block is encoded by the entropy encoding unit 155 and transmitted to the video decoding apparatus.

The inter-prediction unit 124 may perform interpolation on a reference picture or reference block in order to increase prediction accuracy. That is, subsamples between two consecutive integer samples are interpolated by applying filter coefficients to a plurality of consecutive integer samples including the two integer samples. When a process of searching for a block most similar to the current block is performed for the interpolated reference picture, the motion vector can be expressed with precision of decimal units instead of integer sample units. The precision or resolution of the motion vector may be set differently for each unit of a target region to be encoded, for example, a slice, tile, CTU, or CU. When such adaptive motion vector resolution (AMVR) is applied, information on motion vector resolution to be applied to each target region must be signaled for each target region. For example, when the target region is a CU, information on motion vector resolution applied to each CU is signaled. Information on the motion vector resolution may be information indicating the precision of differential motion vectors, which will be described later.

Meanwhile, the inter prediction unit 124 may perform inter prediction using bi-prediction. In the case of bi-directional prediction, two reference pictures and two motion vectors representing positions of blocks most similar to the current block within each reference picture are used. The inter prediction unit 124 selects a first reference picture and a second reference picture from reference picture list 0 (RefPicList0) and reference picture list 1 (RefPicList1), respectively, and searches for a block similar to the current block within each reference picture. A first reference block and a second reference block are generated. Then, a prediction block for the current block is generated by averaging or weighted averaging the first reference block and the second reference block. Further, motion information including information on two reference pictures used to predict the current block and information on two motion vectors is delivered to the encoder 150. Here, reference picture list 0 may include pictures prior to the current picture in display order among restored pictures, and reference picture list 1 may include pictures after the current picture in display order among restored pictures. there is. However, it is not necessarily limited to this, and in order of display, ups and downs pictures subsequent to the current picture may be additionally included in reference picture list 0, and conversely, ups and downs pictures prior to the current picture may be additionally included in reference picture list 1. may also be included.

Various methods may be used to minimize the amount of bits required to encode motion information.

For example, when the reference picture and motion vector of the current block are the same as those of the neighboring block, the motion information of the current block can be delivered to the video decoding apparatus by encoding information capable of identifying the neighboring block. This method is called 'merge mode'.

In the merge mode, the inter prediction unit 124 selects a predetermined number of merge candidate blocks (hereinafter referred to as 'merge candidates') from neighboring blocks of the current block.

Neighboring blocks for deriving merge candidates include a left block (A0), a lower left block (A1), an upper block (B0), and an upper right block (B1) adjacent to the current block in the current picture, as shown in FIG. ), and all or part of the upper left block A2 may be used. Also, a block located in a reference picture (which may be the same as or different from a reference picture used to predict the current block) other than the current picture in which the current block is located may be used as a merge candidate. For example, a block co-located with the current block in the reference picture or blocks adjacent to the co-located block 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 0 vector is added to the merge candidates.

The inter prediction unit 124 constructs a merge list including a predetermined number of merge candidates using these neighboring blocks. Among the merge candidates included in the merge list, a merge candidate to be used as motion information of the current block is selected, and merge index information for identifying the selected candidate is generated. The generated merge index information is encoded by the encoder 150 and transmitted to the video decoding apparatus.

Merge skip mode is a special case of merge mode. After performing quantization, when all transform coefficients for entropy encoding are close to zero, only neighboring block selection information is transmitted without transmitting a residual signal. By using the merge skip mode, it is possible to achieve a relatively high encoding efficiency in low-motion images, still images, screen content images, and the like.

Hereinafter, merge mode and merge skip mode are collectively referred to as merge/skip mode.

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

In the AMVP mode, the inter prediction unit 124 derives predictive motion vector candidates for the motion vector of the current block using neighboring blocks of the current block. Neighboring blocks used to derive predictive motion vector candidates include a left block A0, a lower left block A1, an upper block B0, and an upper right block adjacent to the current block in the current picture shown in FIG. B1), and all or part of the upper left block (A2) may be used. In addition, a block located in a reference picture (which may be the same as or different from the reference picture used to predict the current block) other than the current picture where the current block is located will be used as a neighboring block used to derive motion vector candidates. may be For example, a collocated block co-located with the current block within the reference picture or blocks adjacent to the collocated block may be used. If the number of motion vector candidates is smaller than the preset number according to the method described above, a 0 vector is added to the motion vector candidates.

The inter-prediction unit 124 derives predicted motion vector candidates using the motion vectors of the neighboring blocks, and determines a predicted motion vector for the motion vector of the current block using the predicted motion vector candidates. Then, a differential motion vector is calculated by subtracting the predicted motion vector from the motion vector of the current block.

The predicted motion vector may be obtained by applying a predefined function (eg, median value, average value operation, etc.) to predicted motion vector candidates. In this case, the video decoding apparatus also knows the predefined function. In addition, since a neighboring block used to derive a predicted motion vector candidate is a block that has already been encoded and decoded, the video decoding apparatus also knows the motion vector of the neighboring block. Therefore, the video encoding apparatus does not need to encode information for identifying a predictive motion vector candidate. Therefore, in this case, information on differential motion vectors and information on reference pictures used to predict the current block are encoded.

Meanwhile, the predicted motion vector may be determined by selecting one of the predicted motion vector candidates. In this case, along with information on differential motion vectors and information on reference pictures used to predict the current block, information for identifying the selected predictive motion vector candidate is additionally encoded.

The subtractor 130 subtracts the prediction block generated by the intra prediction unit 122 or the inter prediction unit 124 from the current block to generate a residual block.

The transform unit 140 transforms the residual signal in the residual block having pixel values in the spatial domain into transform coefficients in the frequency domain. The transform unit 140 may transform residual signals in the residual block by using the entire size of the residual block as a transform unit, or divide the residual block into a plurality of subblocks and use the subblocks as a transform unit to perform transformation. You may. Alternatively, the residual signals may be divided into two subblocks, a transform region and a non-transform region, and transform the residual signals using only the transform region subblock as a transform unit. Here, the transformation region subblock may be one of two rectangular blocks having a size ratio of 1:1 based on a horizontal axis (or a vertical axis). In this case, a flag (cu_sbt_flag) indicating that only subblocks have been transformed, directional (vertical/horizontal) information (cu_sbt_horizontal_flag), and/or location information (cu_sbt_pos_flag) are encoded by the entropy encoding unit 155 and signaled to the video decoding device. do. In addition, the size of the transform region subblock may have a size ratio of 1:3 based on the horizontal axis (or vertical axis), and in this case, a flag (cu_sbt_quad_flag) for distinguishing the corresponding division is additionally encoded by the entropy encoder 155 to obtain an image It is signaled to the decryption device.

Meanwhile, the transform unit 140 may individually transform the residual block in the horizontal direction and the vertical direction. For the transformation, various 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 transform set (MTS). The transform unit 140 may select one transform function pair having the highest transform efficiency among the MTS and transform the residual blocks in the horizontal and vertical directions, respectively. Information (mts_idx) on a pair of transform functions selected from the MTS is encoded by the entropy encoding unit 155 and signaled to the video decoding device.

The quantization unit 145 quantizes transform coefficients output from the transform unit 140 using a quantization parameter, and outputs the quantized transform coefficients to the entropy encoding unit 155 . The quantization unit 145 may directly quantize a related residual block without transformation for a certain block or frame. The quantization unit 145 may apply different quantization coefficients (scaling values) according to positions of transform coefficients in the transform block. A quantization matrix applied to the two-dimensionally arranged quantized transform coefficients may be coded and signaled to the video decoding apparatus.

The rearrangement unit 150 may rearrange the coefficient values of the quantized residual values.

The reordering unit 150 may change a 2D coefficient array into a 1D coefficient sequence using coefficient scanning. For example, the reordering unit 150 may output a one-dimensional coefficient sequence by scanning DC coefficients to coefficients in a high frequency region using a zig-zag scan or a diagonal scan. . Depending on the size of the transformation unit and intra prediction mode, instead of zig-zag scan, vertical scan that scans a 2D coefficient array in a column direction and horizontal scan that scans 2D block-shaped coefficients in a row direction may be used. That is, a scan method to be used among zig-zag scan, diagonal scan, vertical scan, and horizontal scan may be determined according to the size of the transform unit and the intra prediction mode.

The entropy encoding unit 155 uses various encoding schemes such as CABAC (Context-based Adaptive Binary Arithmetic Code) and Exponential Golomb to convert the one-dimensional quantized transform coefficients output from the reordering unit 150 to each other. A bitstream is created by encoding the sequence.

In addition, the entropy encoding unit 155 encodes information such as CTU size, CU splitting flag, QT splitting flag, MTT splitting type, and MTT splitting direction related to block splitting so that the video decoding apparatus can divide the block in the same way as the video encoding apparatus. make it possible to divide In addition, the entropy encoding unit 155 encodes information about a prediction type indicating whether the current block is encoded by intra prediction or inter prediction, and encodes intra prediction information (ie, intra prediction) according to the prediction type. mode) or inter prediction information (motion information encoding mode (merge mode or AMVP mode), merge index in case of merge mode, reference picture index and differential motion vector information in case of AMVP mode) are encoded. Also, the entropy encoding unit 155 encodes information related to quantization, that is, information about quantization parameters and information about quantization matrices.

The inverse quantization unit 160 inversely quantizes the quantized transform coefficients output from the quantization unit 145 to generate transform coefficients. The inverse transform unit 165 transforms transform coefficients output from the inverse quantization unit 160 from a frequency domain to a spatial domain to restore a residual block.

The adder 170 restores the current block by adding the restored residual block and the predicted block generated by the predictor 120. Pixels in the reconstructed current block are used as reference pixels when intra-predicting the next block.

The loop filter unit 180 reconstructs pixels in order to reduce blocking artifacts, ringing artifacts, blurring artifacts, etc. caused by block-based prediction and transformation/quantization. perform filtering on The filter unit 180 is an in-loop filter and may include all or part of a deblocking filter 182, a sample adaptive offset (SAO) filter 184, and an adaptive loop filter (ALF) 186. .

The deblocking filter 182 filters the boundary between reconstructed blocks to remove blocking artifacts caused by block-by-block encoding/decoding, and the SAO filter 184 and alf 186 perform deblocking filtering. Additional filtering is performed on the image. The SAO filter 184 and the alf 186 are filters used to compensate for a difference between a reconstructed pixel and an original pixel caused by lossy coding. The SAO filter 184 improves not only subjective picture quality but also coding efficiency by applying an offset in units of CTUs. In contrast, the ALF 186 performs block-by-block filtering. Distortion is compensated for by applying different filters by distinguishing the edge of the corresponding block and the degree of change. Information on filter coefficients to be used for ALF may be coded and signaled to the video decoding apparatus.

The reconstruction block filtered through the deblocking filter 182, the SAO filter 184, and the ALF 186 is 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-prediction of blocks in the picture to be encoded later.

5 is an exemplary block diagram of a video decoding apparatus capable of implementing the techniques of this disclosure. Hereinafter, referring to FIG. 5, a video decoding device and sub-elements of the device will be described.

The image decoding apparatus includes an entropy decoding unit 510, a rearrangement unit 515, an inverse quantization unit 520, an inverse transform unit 530, a prediction unit 540, an adder 550, a loop filter unit 560, and a memory ( 570) may be configured.

Like the image encoding device of FIG. 1 , each component of the image decoding device may be implemented as hardware or software, or a combination of hardware and software. Also, the function of each component may be implemented as software, and the microprocessor may be implemented to execute the software function corresponding to each component.

The entropy decoding unit 510 determines a current block to be decoded by extracting information related to block division by decoding the bitstream generated by the video encoding apparatus, and provides prediction information and residual signals necessary for restoring the current block. extract information, etc.

The entropy decoding unit 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 the determined size. Then, the CTU is divided using the tree structure by determining the CTU as the top layer of the tree structure, that is, the root node, and extracting division information for the CTU.

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

As another example, when a CTU is split using the QTBTTT structure, a CU split flag (split_cu_flag) indicating whether the CU is split is first extracted, and when the corresponding block is split, a first flag (QT_split_flag) is extracted. may be During the splitting process, each node may have zero or more iterative MTT splits after zero or more repetitive QT splits. For example, the CTU may immediately undergo MTT splitting, or conversely, only QT splitting may occur multiple times.

As another example, when a CTU is split using a QTBT structure, a first flag (QT_split_flag) related to QT splitting is extracted and each node is split into four nodes of a lower layer. And, for a node corresponding to a leaf node of QT, a split flag (split_flag) indicating whether to further split into BTs and split direction information are extracted.

Meanwhile, when the entropy decoding unit 510 determines a current block to be decoded by using tree structure partitioning, it extracts information about a prediction type indicating whether the current block is intra-predicted or inter-predicted. When the prediction type information indicates intra prediction, the entropy decoding unit 510 extracts syntax elements for intra prediction information (intra prediction mode) of the current block. When the prediction type information indicates inter prediction, the entropy decoding unit 510 extracts syntax elements for the inter prediction information, that is, information indicating a motion vector and a reference picture to which the motion vector refers.

In addition, the entropy decoding unit 510 extracts quantization-related information and information about quantized transform coefficients of the current block as information about the residual signal.

The reordering unit 515 converts the sequence of 1-dimensional quantized transform coefficients entropy-decoded in the entropy decoding unit 510 into a 2-dimensional coefficient array (ie, in the reverse order of the coefficient scanning performed by the image encoding apparatus). block) can be changed.

The inverse quantization unit 520 inverse quantizes the quantized transform coefficients and inverse quantizes the quantized transform coefficients using a quantization parameter. The inverse quantization unit 520 may apply different quantization coefficients (scaling values) to the two-dimensionally arranged quantized transform coefficients. The inverse quantization unit 520 may perform inverse quantization by applying a matrix of quantization coefficients (scaling values) from the image encoding device to a 2D array of quantized transformation coefficients.

The inverse transform unit 530 inversely transforms the inverse quantized transform coefficients from the frequency domain to the spatial domain to restore residual signals, thereby generating a residual block for the current block.

In addition, when the inverse transform unit 530 inverse transforms only a partial region (subblock) of a transform block, a flag (cu_sbt_flag) indicating that only a subblock of the transform block has been transformed, and direction information (vertical/horizontal) information (cu_sbt_horizontal_flag) of the transform block ) and/or the location information (cu_sbt_pos_flag) of the subblock, and inversely transforms the transform coefficients of the corresponding subblock from the frequency domain to the spatial domain to restore the residual signals. By filling , the final residual block for the current block is created.

In addition, when the MTS is applied, the inverse transform unit 530 determines transform functions or transform matrices to be applied in the horizontal and vertical directions, respectively, using MTS information (mts_idx) signaled from the video encoding device, and uses the determined transform functions. Inverse transform is performed on the transform coefficients in the transform block in the horizontal and vertical directions.

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

The intra prediction unit 542 determines the intra prediction mode of the current block among a plurality of intra prediction modes from the syntax element for the intra prediction mode extracted from the entropy decoding unit 510, and references the current block according to the intra prediction mode. The current block is predicted using pixels.

The inter prediction unit 544 determines the motion vector of the current block and the reference picture referred to by the motion vector by using the syntax element for the inter prediction mode extracted from the entropy decoding unit 510, and converts the motion vector and the reference picture. to predict the current block.

The adder 550 restores the current block by adding the residual block output from the inverse transform unit and the prediction block output from the inter prediction unit or intra prediction unit. Pixels in the reconstructed current block are used as reference pixels when intra-predicting a block to be decoded later.

The loop filter unit 560 may include a deblocking filter 562, an SAO filter 564, and an ALF 566 as in-loop filters. The deblocking filter 562 performs deblocking filtering on boundaries between reconstructed blocks in order to remove blocking artifacts generated by block-by-block decoding. The SAO filter 564 and the ALF 566 perform additional filtering on the reconstructed block after deblocking filtering to compensate for the difference between the reconstructed pixel and the original pixel caused by lossy coding. ALF filter coefficients are determined using information on filter coefficients decoded from the non-stream.

The reconstruction block filtered through the deblocking filter 562, the SAO filter 564, and the ALF 566 is stored in the memory 570. When all blocks in one picture are reconstructed, the reconstructed picture is used as a reference picture for inter-prediction of blocks in the picture to be encoded later.

This embodiment relates to encoding and decoding of images (video) as described above. More specifically, in order to improve video encoding efficiency and video quality, a residual frame is generated from a reconstructed frame using a deep learning model, and the generated residual frame is used as a linear model. Provided is a video coding method and apparatus for improving the performance of an in-loop filter by approximating an original residual frame by applying a linear model.

The following embodiments may be commonly applied to the loop filter unit 180 in the video encoding apparatus and the loop filter unit 560 in the video decoding apparatus. In addition, the following embodiments may be commonly applied to a part using deep learning technology of an image encoding device and an image decoding device.

In the following description, the term 'target block' may be used in the same meaning as the current block or coding unit (CU, Coding Unit) as described above, or may mean a partial region of the coding unit. .

6 is an exemplary diagram illustrating a picture quality improving apparatus according to an embodiment of the present disclosure.

The picture quality improving apparatus according to the present embodiment generates a residual frame from a restored frame using a deep learning model, applies the generated residual frame to a linear model to approximate the original residual frame, and uses the approximated original residual frame to improve the image quality. Create an improved recovery frame. The image quality improving device includes all or part of the first enhancer 602 , the subtracter 604 , the second enhancer 606 , and the adder 608 . In addition, the first improvement unit 602 includes an improvement model that is a deep learning model, and the second improvement unit 606 includes a linear model.

When the picture quality improving device according to the present embodiment is included in the video encoding device, components included in the picture quality improving device are not necessarily limited to the example of FIG. 1 . For example, the picture quality improving apparatus may additionally include a training unit (not shown) for training the enhancement model, or may be implemented in a form interlocked with an external training unit.

The first enhancement unit 602 obtains the restored frame x _rec [i, j] as an input, inputs the restored frame x _rec [i, j] to the improved model, and approximates the original frame x _ori [i, j] as an output Generates x _nn [i,j]. Here, [i,j] represents the position of a pixel in a frame. The improved model may generate an approximate output x _nn using a plurality of layers. Here, the plurality of layers may be convolutional layers suitable for deep learning-based video signal processing.

Subtractor 604 subtracts the reconstructed frame x _rec [i, j] from the approximate output x _nn [i, j] to produce a residual frame x _res [i, j], as shown in equation (1).

Thus, the residual frame x _res approximates the difference between the original frame x _ori and the reconstructed frame x _rec . Hereinafter, this difference is referred to as 'original residual frame x _{o_res} '. The original residual frame x _{o_res} [i, j] can be expressed as in Equation 2.

The second refiner 606 approximates the original residual frame x _{o_res} once more from the residual frame x _res using a linear model. When α and β are the parameters representing the linear relation provided by the linear model, the linearly approximated residual frame x' _res [i,j] can be expressed as in Equation 3.

Hereinafter, the residual frame x _res is used interchangeably with the 'first residual frame', and the linearly approximated residual frame x' _res is interchangeably used with the 'second residual frame'.

As shown in Equation 4, the adder 608 adds the second residual frame x' _res [i,j] and the reconstructed frame x _rec [i,j] to obtain an improved reconstructed frame x' _rec [i,j] generate

From Equations 3 and 4, the second residual frame x' _res [i,j] can be expressed as Equation 5.

Meanwhile, the training unit may train the improved model by using the original residual frame x _{o_res} as a target so that the improved model can learn to approximate the original frame x _ori . The training unit may update parameters included in the plurality of layers using a loss function as shown in Equation 6.

Here, ?·? denotes the L2 metric. Meanwhile, as shown in Equation 6, the loss function of the improvement model is expressed as an L2 metric, but is not necessarily limited thereto. Any metric that can represent the distance between the first residual frame x _res and the original residual frame x _{o_res} can be used as the loss function.

Parameters α and β of the linear model may be calculated using the first residual frame x _res [i,j] and the second residual frame _x'res [i,j]. In this case, since the second residual frame x' _res [i, j] approximates the value not yet determined or the original residual frame, the original residual frame x _{o_res} [i, j] may be used instead. Utilizing the linear least square equation, α and β can be estimated as shown in Equation 7.

Here, K represents the total number of pixels included in the frame. That is, to calculate α and β, [i,j] can be replaced by [k]. Thus, the linear model provides a linear relationship between pixels in the first residual frame x _res and pixels in the original residual frame x _{o_res} .

Meanwhile, when one of α and β is set to 0, a non-zero parameter may be estimated according to Equation 7. Thereafter, the second improver 606 may generate a second residual frame x' _res [i, j] by using Equation 3 including one estimated parameter.

7 is an exemplary diagram illustrating a picture quality improving apparatus according to another embodiment of the present disclosure.

The picture quality improving apparatus according to the present embodiment may enhance the first residual frame x _res by using an embedding vector of video encoding/decoding information. The picture quality improving apparatus may additionally include an embedding vector generator 702 and a matrix multiplier 704 .

The embedding vector generator 702 generates an embedding vector λ _P from video encoding/decoding information P using an embedding function e(·). Here, the embedding function is a deep learning model composed of a fully-connected layer or a convolutional layer.

The encoding/decoding information P includes a quantization parameter, a Lagrange constant used in the bit rate distortion optimization process, a temporal layer within a group of pictures (GOP), and a type of frame (I/P /B, Intra/Predictive/Bi-Predictive), etc. may be used. Alternatively, a combination of all or part of these may be used as the encoding/decoding information P.

The matrix multiplier 704 matrix-multiplies the embedding vector λ _P and the first residual frame x _res to generate a complemented residual frame x _{h_res} [i,j] having the same size as x _res . Here, the matrix multiplication may be, for example, component-wise multiplication between the embedding vector and rows in the first residual frame, for which the size of the embedding vector is the same as the size of the row in the first residual frame. , an embedding function can be defined. As another example, matrix multiplication may be element-by-element multiplication between an embedding vector and columns in the first residual frame. To this end, an embedding function may be defined such that the size of the embedding vector is equal to the size of a column in the first residual frame. there is.

The second refiner 606 approximates the original residual frame x _{o_res} once more from the augmented residual frame x _{h_res} [i,j] using a linear model. For the parameters α and β representing the linear model, the linearly approximated residual frame x' _res [i, j] can be expressed as Equation 8.

In this case, parameters α and β of the linear model may be estimated as shown in Equation 9 using the original residual frame x _{o_res} [i, j] and the enhanced residual frame x _{h_res} [i, j].

As an example, the video encoding apparatus may derive the parameters of the linear model, encode them into a bitstream, and transmit the data to the video decoding apparatus. In this case, whether or not to apply picture quality improvement may be determined at the frame, picture, subpicture, slice, tile, block, and pixel levels. In the case of block units, whether to apply picture quality improvement may be determined based on CTUs and CU/Prediction Units (CUs/PUs). Alternatively, whether to apply picture quality improvement may be determined based on the same size as the sub-CU. Whether or not to apply picture quality improvement may be determined based on a size such as a tile or a subpicture as a set of blocks.

As another example, the video encoding apparatus may signal to the video decoding apparatus whether or not the image quality is improved by using a flag. An image encoding device may receive this flag from a higher level, and an image decoding device may obtain this flag by decoding a bitstream. On the other hand, if the method of encoding the parameter information regardless of the application unit is more damaging in terms of bit rate distortion than the existing method of transmitting only the restored signal, the picture quality improvement method may not be applied according to the flag value.

As another example, in addition to the method of transmitting parameter values of a linear model for all application units, parameter values according to video encoding/decoding information may be statistically utilized. For example, when the quantization parameter is 22, parameter values of a linear model that frequently occur may be pre-calculated, and then parameter values may be preset using the calculated values. Alternatively, after generating a lookup table based on statistics of parameter values according to encoding/decoding information, the lookup table may be used. At this time, after one or more lookup tables are created according to the quantization parameter value, frame type, temporal ID in the GOP, size of the applied block, whether the block is intra/inter coded, etc., it can be used when the picture quality improvement method is applied. there is.

As another example, a plurality of (α, β) pairs and corresponding indexes may be set in advance based on statistics of parameter values according to encoding/decoding information. The video encoding apparatus may signal an index, and the video decoding apparatus may obtain predefined α and β using the index. In this case, when one of α or β is fixed to 0, an index may be signaled for a parameter value other than 0.

Meanwhile, the picture quality improving device may be one filter included in the loop filter unit 180 in the video encoding device or the loop filter unit 560 in the video decoding device. Accordingly, the restored frame x _rec is a frame obtained by reconstructing the original frame, and any frame previously restored by the video encoding device or the video decoding device may be used. For example, the restored frame x _rec is a signal stored in a decoded picture buffer (DPB) in the memory 190 or 570, an output of the deblocking filter 182 or 562, an output of the SAO filter 184 or 564, or a bilateral filter (BF) It may be one of the output of , and the output of the ALF (186, 566). Here, BF is one of the in-loop filters and can be applied to improve the picture quality of the restored frame. In addition, the signal stored in the DPB is defined as an output of the adders 170 and 550, that is, a restored signal corresponding to the sum of the prediction signal and the inverse transform signal. Meanwhile, as illustrated in FIGS. 1 and 5 , the DBP may store reconstructed pictures generated by passing through all in-loop filters.

As another example, in addition to the in-loop filter, x _rec may be an inter-prediction signal or an intra-prediction signal, and improvement of the prediction signal, improvement of picture quality after compression, and the like may be performed by the picture quality improvement method according to the present embodiment.

In the examples of FIGS. 6 and 7 , x _nn according to Equation 1 is an output of an enhancement model that is a deep learning model. However, when a deep learning model is not used, x _nn is the output of the deblocking filter (182, 562), the output of the SAO filter (184, 564), the output of the bilateral filter, and the output of the ALF (186, 566) can be one of

8 is an exemplary diagram illustrating a picture quality improving apparatus according to another embodiment of the present disclosure.

For example, as illustrated in FIG. 8, x _rec is used as an output of the SAO 564 and an input of the ALF 566, and x _nn may be used as an output of the ALF 566. Then, x _nn may be used to generate the improved recovery signal x' _rec . The picture quality improvement apparatus illustrated in FIG. 8 is configured based on in-loop filters in the video decoding apparatus. It is not necessarily limited to this. For example, the picture quality improving device may be similarly configured based on in-loop filters in the video encoding device.

Hereinafter, a picture quality improvement method using a linear model will be described using FIGS. 9 and 10 .

9 is a flowchart illustrating a method for improving an image quality according to an embodiment of the present disclosure.

The picture quality improving device obtains a restored frame (S900).

The picture quality improving apparatus generates an output approximating the original frame by inputting the restored frame to the deep learning-based enhancement model (S902). Here, the restored frame is a frame obtained by restoring an original frame, and is previously restored by an image encoding device or an image decoding device. For example, the restored frame includes signals stored in DPBs in the memories 190 and 570, outputs of the deblocking filters 182 and 562, outputs of the SAO filters 184 and 564, outputs of the BFs, and ALFs 186 and 566. It can be one of the outputs.

The picture quality improving apparatus generates a first residual frame by subtracting the restored frame from the output of the improvement model (S904). The first residual frame approximates the original residual frame, and the original residual frame is the difference between the original frame and the reconstructed frame.

The enhancement model is a deep learning model that includes multiple layers and is trained to learn to approximate the original frame. An improved model may be trained using a loss function based on the difference between the first residual frame and the original residual frame, as shown in Equation (6).

The picture quality improving apparatus generates a second residual frame by inputting the first residual frame to the linear model (S906). Here, the linear model includes parameters α and β representing a linear relationship between the first residual frame and the second residual frame.

Parameters of the linear model may be calculated using the first residual frame and the second residual frame. At this time, the second residual frame is a value that has not yet been determined, but since it approximates the original residual frame, the original residual frame may be used instead. Parameters of the linear model may be calculated as shown in Equation 7 utilizing the linear least squares equation. The linear model may provide a linear relationship between pixel values in the first residual frame and pixel values in the original residual frame.

The picture quality improving apparatus generates an improved restored frame by adding the second residual frame and the restored frame (908).

10 is a flowchart illustrating a method for improving an image quality according to an embodiment of the present disclosure.

The picture quality improving device obtains a restored frame (S1000).

The picture quality improving apparatus generates an output approximating the original frame by inputting the restored frame to the deep learning-based enhancement model (S1002). Here, the restored frame is a frame obtained by restoring an original frame, and is previously restored by an image encoding device or an image decoding device.

The picture quality improving apparatus generates a first residual frame by subtracting the restored frame from the output of the improvement model (S1004).

The picture quality improving device acquires encoding/decoding information (S1006). Here, as the encoding/decoding information, a quantization parameter, a Lagrange constant used in a bit rate distortion optimization process, a temporal layer in a GOP, a frame type (I/P/B), and the like may be used. Alternatively, a combination of all or part of these may be used as encoding/decoding information.

The picture quality improving apparatus generates an embedding vector by inputting encoding/decoding information to a deep learning-based embedding function (S1008). The embedding function is a deep learning model composed of fully connected layers or convolutional layers.

The picture quality improving apparatus generates an enhancement residual frame by matrix-multiplying the embedding vector and the first residual frame (S1010). Here, matrix multiplication may be, for example, element-by-element multiplication between the embedding vector and rows in the first residual frame. As another example, matrix multiplication may be element-wise multiplication between the embedding vector and columns in the first residual frame,

The picture quality improving apparatus generates a second residual frame by inputting the augmented residual frame into the linear model (S1012). Here, the linear model includes parameters α and β representing a linear relationship between the reinforcement residual frame and the second residual frame. Parameters of the linear model can be calculated as shown in Equation 9 using the original residual frame and the augmented residual frame.

The picture quality improving apparatus generates an improved and restored frame by adding the second residual frame and the restored frame (S1014).

In the flow chart/timing diagram of the present specification, it is described that each process is sequentially executed, but this is merely an example of the technical idea of one embodiment of the present disclosure. In other words, those skilled in the art to which an embodiment of the present disclosure belongs may change and execute the order described in the flowchart/timing diagram within the range that does not deviate from the essential characteristics of the embodiment of the present disclosure, or one of each process Since the above process can be applied by performing various modifications and variations in parallel, the flow chart/timing chart is not limited to a time-series sequence.

In the above description, it should be understood that the exemplary embodiments may be implemented in many different ways. Functions or methods described in one or more examples may be implemented in hardware, software, firmware, or any combination thereof. It should be understood that the functional components described in this specification have been labeled "...unit" to particularly emphasize their implementation independence.

Meanwhile, various functions or methods described in this embodiment may be implemented as instructions stored in a non-transitory storage medium that can be read and executed by one or more processors. Non-transitory recording media include, for example, all types of recording devices in which data is stored in a form readable by a computer system. For example, the non-transitory recording medium includes storage media 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).

The above description is merely an example of the technical idea of the present embodiment, and various modifications and variations can be made to those skilled in the art without departing from the essential characteristics of the present embodiment. Therefore, the present embodiments are not intended to limit the technical idea of the present embodiment, but to explain, and the scope of the technical idea of the present embodiment is not limited by these embodiments. The scope of protection of this embodiment should be construed according to the claims below, and all technical ideas within the scope equivalent thereto should be construed as being included in the scope of rights of this embodiment.

180: loop filter unit
560: loop filter unit
602: first improvement section
606: second improvement section
702: embedding vector generation unit

Claims (17)

A method for improving the quality of a reconstructed frame, performed by a computing device, comprising:
Obtaining the restored frame, and generating an output approximating an original frame by inputting the restored frame to a deep learning-based improvement model, wherein the restored frame comprises reconstructing the original frame frame, previously restored by the computing device;
generating a first residual frame by subtracting the restored frame from the output of the improved model;
generating a second residual frame by inputting the first residual frame into a linear model, wherein the linear model includes parameters indicating a linear relationship between the first residual frame and the second residual frame; and
generating an improved reconstruction frame by adding the second residual frame and the reconstruction frame;
Characterized in that, the method comprising a. According to claim 1,
The first residual frame,
Approximating an original residual frame, wherein the original residual frame is a difference between the original frame and the reconstructed frame. According to claim 2,
The improvement model,
A deep learning model comprising a plurality of layers, characterized in that it is trained using a loss function based on a difference between the first residual frame and the original residual frame. According to claim 1,
The restoration frame,
One of a reconstruction signal stored in a decoded picture buffer (DPB), an output of a deblocking filter, an output of a sample adaptive offset (SAO) filter, and an output of an adaptive loop filter (ALF), wherein the reconstruction signal is a prediction signal and an inverse transformation signal. characterized in that it is a sum. According to claim 1,
The method further comprising signaling a flag indicating whether or not the method of improving picture quality is applied. According to claim 2,
The parameters of the linear model are,
characterized in that the estimation is made utilizing a linear least square equation based on pixel values in the first residual frame and pixel values in the original residual frame. According to claim 1,
The parameters of the linear model are,
Characterized in that, after being induced by a video encoding device and encoded into a bitstream, it is transmitted to a video decoding device. According to claim 1,
The parameters of the linear model are,
A method using preset values, characterized in that the preset values are statistically frequently occurring values based on encoding and decoding information. According to claim 1,
The parameters of the linear model are,
Parameter values of the linear model and a corresponding index are set in advance, the video encoding device signals the index to the video decoding device, and the video decoding device acquires the parameter values set in advance using the index. characterized, how. In the picture quality improving device included in the video decoding device,
A first improvement unit that includes a deep learning-based improvement model, obtains a reconstructed frame, and generates an output approximating an original frame by inputting the reconstructed frame to the improvement model, wherein , The restored frame is a frame obtained by restoring the original frame, and is previously restored by the video decoding apparatus;
a subtractor configured to subtract the reconstructed frame from the output of the improved model to generate a first residual frame;
A second improvement unit including a linear model and generating a second residual frame by inputting the first residual frame to the linear model, wherein the linear model is a linear relationship between the first residual frame and the second residual frame including parameters representing; and
An adder generating an improved reconstruction frame by adding the second residual frame and the reconstruction frame
Characterized in that it comprises a picture quality improving device. According to claim 10,
The first residual frame,
Approximate an original residual frame, wherein the original residual frame is a difference between the original frame and the restored frame. According to claim 11,
The improvement model,
A deep learning model including a plurality of layers, characterized in that it is trained using a loss function based on a difference between the first residual frame and the original residual frame. According to claim 11,
The parameters of the linear model are,
characterized in that the estimation is performed using a linear least square equation based on pixel values in the first residual frame and pixel values in the original residual frame. A method for improving the quality of a reconstructed frame, performed by a computing device, comprising:
Obtaining the restored frame and generating an output approximating the original frame by inputting the restored frame to a deep learning-based improvement model, wherein the restored frame is a frame obtained by reconstructing the original frame; previously restored by the computing device;
generating a first residual frame by subtracting the restored frame from the output of the improved model;
Obtaining encoding/decoding information, and generating an embedding vector by inputting the encoding/decoding information into a deep learning-based embedding function, wherein the encoding/decoding information includes a quantization parameter and bit rate distortion. At least one of a Lagrange constant used in an optimization process, a temporal layer in a group of pictures (GOP), and a type of frame;
generating a complemented residual frame by matrix multiplication of the embedding vector and the first residual frame;
generating a second residual frame by inputting the augmented residual frame into a linear model, wherein the linear model includes parameters indicating a linear relationship between the augmented residual frame and the second residual frame; and
generating an improved reconstruction frame by adding the second residual frame and the reconstruction frame;
Characterized in that, the method comprising a. According to claim 14,
The first residual frame,
Approximating an original residual frame, wherein the original residual frame is a difference between the original frame and the reconstructed frame. According to claim 15,
The improvement model,
A deep learning model comprising a plurality of layers, characterized in that it is trained using a loss function based on a difference between the first residual frame and the original residual frame. According to claim 15,
The parameters of the linear model are,
characterized in that the estimation is made utilizing a linear least squares equation based on pixel values in the enhanced residual frame and pixel values in the original residual frame.

Images