KR102580910B1 - Motion Compensation Method and Apparatus Using Bi-directional Optical Flow - Google Patents

Abstract

The present invention discloses an adaptive bidirectional optical flow (BIO) estimation method for inter-screen prediction correction during the video encoding process. The purpose of the present invention is to reduce the complexity and/or cost of bidirectional optical flow at the pixel level or subblock level.

Description

Motion Compensation Method and Apparatus Using Bi-directional Optical Flow}

The present invention relates to video encoding or decoding. More specifically, it relates to bidirectional optical flow for correction of motion compensation.

The content described in this section simply provides background information for this embodiment and does not constitute prior art.

As far as video coding is concerned, compression is achieved by exploiting data redundancy in spatial and temporal dimensions. Spatial redundancy reduction is largely achieved by transform coding. Temporal redundancy is reduced through predictive coding. Observing that temporal correlation is maximized along the motion trajectory, motion compensation prediction is used for this purpose. The main goal of motion estimation in this context is not to find 'real' motion in the scene, but to maximize compression efficiency. In other words, the motion vector must provide an accurate prediction of the signal. Additionally, because motion information must be transmitted as overhead in a compressed bitstream, compressed representation must be possible. Efficient motion estimation is important to achieve high compression in video coding.

In video sequences, movement is an important source of information. Motion occurs not only due to object but also camera movement. Apparent motion, also known as optical flow, captures spatio-temporal variations in pixel intensity in an image sequence.

Bi-directional Optical Flow (BIO) is a motion estimation/compensation technology disclosed in JCTVC-C204 and VCEG-AZ05 BIO, which performs sample-level motion adjustment ( motion refinement). The bidirectional optical flow estimation method currently under discussion has the advantage of enabling detailed correction of motion vector information, but has the disadvantage of requiring much higher computational complexity than traditional bidirectional prediction for detailed motion vector information correction.

Non-patent Document 1: JCTVC-C204 (E. Alshina, et al., Bi-directional optical flow, Joint Collaborative Team on Video Coding (JCT-VC) of ITU-T SG 16 WP 3 and ISO/IEC JTC 1/SC 29/WG 11, 3rd Meeting: Guangzhou, CN, 7-15 October, 2010)

Non-patent Document 2: VCEG-AZ05 (E. Alshina, et al., Known tools performance investigation for next generation video coding, ITU-T SG 16 Question 6, Video Coding Experts Group (VCEG), 52nd Meeting: 19-26 June 2015, Warsaw, Poland)

The purpose of the present invention is to reduce image quality degradation while reducing the computational complexity of bi-directional optical flow (BIO).

According to one aspect of the present invention, generating a first reference block using a first motion vector referring to a first reference picture, and generating a second reference block using a second motion vector referring to a second reference picture. ; calculating texture complexity of the current block using the first reference block and the second reference block; And generate a prediction block of the current block, selectively based on the texture complexity, using the first and second reference blocks according to the BIO process, or generate a prediction block of the current block using the BIO process. Provides a motion compensation method using bi-directional optical flow (BIO), comprising generating a prediction block of the current block using the first and second reference blocks without applying the motion compensation method. do.

According to another aspect of the present invention, a first reference block is generated using a first motion vector referring to a first reference picture, and a second reference block is generated using a second motion vector referring to a second reference picture. Block generation unit; a skip decision unit that calculates texture complexity of the current block using the first reference block and the second reference block, and determines whether to skip the BIO process by comparing the texture complexity with a threshold; And selectively based on the decision of the skip decision unit, generate a prediction block of the current block using the first and second reference blocks according to the BIO process, or generate the first and second reference blocks without applying the BIO process. 2. A motion compensation device using bi-directional optical flow (BIO) is provided, which includes a prediction block generator that generates a prediction block of the current block using a reference block.

1 is an exemplary block diagram of an image encoding device according to an embodiment of the present invention.
Figure 2 is an example of the surrounding blocks of the current block,
3 is an exemplary block diagram of an image decoding device according to an embodiment of the present invention;
Figure 4 is a reference diagram for explaining the basic concept of BIO;
Figure 5 is an example of the shape of a mask centered on the current pixel in pixel-based BIO.
Figure 6 is an example diagram to explain setting the brightness value and gradient for pixels outside the reference block within the mask in a per-pad manner;
Figure 7 is an example of the form of a mask centered on a subblock in subblock-based BIO.
Figure 8 is an example to explain applying a mask for each pixel in subblock-based BIO;
Figure 9 is an example of another mask shape centered on a subblock in subblock-based BIO.
Figure 10 is a block diagram showing the configuration of a device that performs motion compensation by selectively applying a BIO process according to an embodiment of the present invention;
Figure 11 is an example of a process for performing motion compensation by selectively applying a BIO process according to the texture complexity of the current block according to an embodiment of the present invention;
Figure 12 is another example of the process of performing motion compensation by selectively applying the BIO process according to the texture complexity of the current block according to an embodiment of the present invention.
Figure 13 is another example of the process of performing motion compensation by selectively applying the BIO process according to the texture complexity of the current block according to an embodiment of the present invention.
Figure 14 is an example of a process for performing motion compensation by selectively applying a BIO process according to the size of the current block and the encoding mode of the motion vector according to an embodiment of the present invention;
Figure 15 is an example diagram showing a process of performing motion compensation by selectively applying a BIO process according to CVC conditions and BCC conditions according to an embodiment of the present invention;
Figure 16 is an example diagram showing the process of performing motion compensation by selectively applying the BIO process according to the motion vector variance of neighboring blocks according to this embodiment.

Hereinafter, some embodiments of the present invention will be described in detail through illustrative drawings. In adding identification codes to components in each drawing, it should be noted that the same components are given the same codes as much as possible even if they are shown in different drawings. Additionally, in describing the present invention, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the present invention, the detailed description will be omitted.

The techniques of this disclosure generally relate to reducing the complexity and/or cost of bi-directional optical flow (BIO) techniques. BIO can be applied during motion compensation. In general, BIO calculates the motion vector for every pixel or subblock in the current block through optical flow, and updates the predicted value of the pixel or subblock based on the calculated motion vector value of each pixel or subblock. It is used to

1 is an example block diagram of a video encoding device that can implement the techniques of the present disclosure.

The image encoding device includes a block division unit 110, a prediction unit 120, a subtractor 130, a transform unit 140, a quantization unit 145, an encoder 150, an inverse quantization unit 160, and an inverse transform unit ( 165), an adder 170, a filter unit 180, and a memory 190. Each component of an image encoding device may be implemented as a hardware chip, or may be implemented as software and one or more microprocessors may be implemented to execute software functions corresponding to each component.

The block division unit 110 divides each picture constituting the image into a plurality of CTUs (Coding Tree Units) and then recursively divides the CTUs using a tree structure. A leaf node in the tree structure becomes a CU (coding unit), the basic unit of encoding. The tree structure is a QuadTree (QT) in which a parent node (or parent node) is divided into four child nodes (or child nodes) of the same size, or a QT structure and a parent node is divided into two child nodes. A QTBT (QuadTree plus BinaryTree) structure that combines the BinaryTree (BT) structure can be used. In other words, QTBT can be used to divide a CTU into multiple CUs.

In the QTBT (QuadTree plus BinaryTree) structure, the CTU can first be divided into a QT structure. Quadtree splitting can be repeated until the size of the splitting block reaches the minimum block size (MinQTSize) of a leaf node allowed in QT. If the leaf nodes of the quadtree are no larger than the maximum block size (MaxBTSize) of the root node allowed in BT, they can be further partitioned into the BT structure. In BT, multiple split types may exist. For example, in some examples, there may be two types: a type that horizontally divides the block of the node into two blocks of the same size (i.e., symmetric horizontal splitting) and a type that divides it vertically (i.e., symmetric vertical splitting). In addition, there may be an additional type that divides the block of the corresponding node into two asymmetric blocks. The asymmetric form may include dividing the block of the corresponding node into two rectangular blocks with a size ratio of 1:3, or may include dividing the block of the corresponding node diagonally.

Segmentation information generated by the block division unit 110 by dividing the CTU using the QTBT structure is encoded by the encoder 150 and transmitted to the video decoding device.

A CU can have various sizes depending on the QTBT division from the CTU. Hereinafter, the block corresponding to the CU (i.e., leaf node of QTBT) to be encoded or decoded is referred to as the 'current block'.

The prediction unit 120 predicts the 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. Prediction of the current block is typically performed using intra prediction techniques (using data from the picture containing the current block) or inter prediction techniques (using data from pictures coded before the picture containing the current block). It can be done. Inter prediction includes both one-way prediction and two-way prediction.

For each inter predicted block, a set of motion information may be available. One set of motion information may include motion information for forward and backward prediction directions. Here, the forward and backward prediction directions are the two prediction directions of bi-directional prediction mode, and the terms “forward” and “backward” do not necessarily have geometric meaning. Instead, they generally correspond to whether the reference picture will be displayed before (“reverse”) or after (“forward”) the current picture. In some examples, the “forward” and “backward” prediction directions may correspond to reference picture list 0 (RefPicList0) and reference picture list 1 (RefPicList1) of the current picture.

For each prediction direction, motion information includes a reference index and a motion vector. The reference index can be used to identify a reference picture in the current reference picture list (RefPicList0 or RefPicList1). A motion vector has horizontal (x) and vertical (y) components. In general, the horizontal component represents the horizontal displacement within the reference picture relative to the position of the current block in the current picture, which is needed to locate the x coordinate of the reference block. The vertical component represents the vertical displacement within the reference picture relative to the position of the current block, which is needed to locate the y coordinate of the reference block.

The inter prediction unit 124 generates a prediction block for the current block through a motion compensation process. The block most similar to the current block is searched within a reference picture that has been encoded and decoded before the current picture, and a prediction block for the current block is generated using the searched block. Then, a motion vector corresponding to the displacement between the current block in the current picture and the prediction block in the reference picture is generated. Typically, motion estimation is performed on the luma component, and motion vectors calculated based on the luma component are used for both the luma component and the chroma component. Motion information including information about reference pictures and motion vectors used to predict the current block is encoded by the encoder 150 and transmitted to the video decoding device.

Meanwhile, in the case of bi-prediction, the inter prediction unit 124 selects the first reference picture and the second reference picture from reference picture list 0 and reference picture list 1, respectively, and searches for a block similar to the current block within each reference picture. Thus, a first reference block and a second reference block are created. Then, the first reference block and the second reference block are averaged or weighted to generate a prediction block for the current block. Then, motion information including information about the two reference pictures used to predict the current block and information about the two motion vectors is transmitted to the encoder 150. Here, the two motion vectors are a first motion vector corresponding to the displacement between the position of the current block in the current picture and the position of the first reference block in the first reference picture (i.e., a motion vector referring to the first reference picture) and a second motion vector (i.e., a motion vector referring to the second reference picture) corresponding to the displacement between the position of the current block in the current picture and the position of the second reference block in the second reference picture.

Additionally, the inter prediction unit 124 may perform the bi-directional optical flow (BIO) process of the present disclosure to generate a prediction block of the current block using bi-directional prediction. In other words, after determining bidirectional motion vectors for the current block, the inter prediction unit 124 can generate a prediction block for the current block using motion compensation according to the BIO process on a per-image pixel or sub-block basis. there is. In other examples, one or more other units of the encoding device may additionally be involved in performing the BIO process of the present disclosure. Additionally, since the BIO process is performed by applying an explicit equation using previously decoded information shared between the encoding device and the decoding device, signaling of additional information for the BIO process is not required.

Whether to apply the BIO process when compensating for motion using bidirectional prediction can be determined in various ways. Details on the BIO process and whether to apply the BIO process in the motion compensation process will be described later with reference to the drawings in FIG. 4 and below.

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

For example, if the reference picture and motion vector of the current block are the same as the reference picture and motion vector of the neighboring block, the motion information of the current block can be transmitted to the decoding device by encoding information that can identify 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.

As shown in Figure 2, the surrounding blocks for deriving merge candidates include the left block (L), top block (A), top right block (AR), and bottom left block (BL) adjacent to the current block in the current picture. ), all or part of the upper left block (AL) can be used. Additionally, a block located within a reference picture (which may be the same or different from the reference picture used to predict the current block) rather than the current picture where the current block is located may be used as a merge candidate. For example, a block co-located with the current block within the reference picture or blocks adjacent to the co-located block may be additionally used as merge candidates.

The inter prediction unit 124 uses these neighboring blocks to construct a merge list including a predetermined number of merge candidates. A merge candidate to be used as motion information of the current block is selected from among the merge candidates included in the merge list, and merge index information is generated to identify the selected candidate. The generated merge index information is encoded by the encoder 150 and transmitted to the decoding device.

Another way to encode motion information is to encode differential motion vectors.

In this method, the inter prediction unit 124 uses neighboring blocks of the current block to derive predicted motion vector candidates for the motion vector of the current block. The surrounding blocks used to derive predicted motion vector candidates include the left block (L), top block (A), top right block (AR), and bottom left block adjacent to the current block in the current picture shown in FIG. 2. BL), or all or part of the upper left block (AL) can be used. Additionally, a block located within a reference picture (which may be the same or different from the reference picture used to predict the current block) rather than the current picture where the current block is located will be used as a surrounding block used to derive prediction motion vector candidates. It may be possible. For example, a block co-located with the current block within the reference picture or blocks adjacent to the co-located block may be used.

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, the predicted motion vector is subtracted from the motion vector of the current block to calculate the differential motion vector.

The predicted motion vector can be obtained by applying a predefined function (eg, median, average value calculation, etc.) to the predicted motion vector candidates. In this case, the video decoding device also knows the predefined function. In addition, since the neighboring blocks used to derive predicted motion vector candidates are blocks for which encoding and decoding have already been completed, the video decoding device also already knows the motion vectors of the neighboring blocks. Therefore, the video encoding device does not need to encode information to identify the predicted motion vector candidate. Therefore, in this case, information about the differential motion vector and information about the reference picture 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, information for identifying the selected prediction motion vector candidate is additionally encoded, along with information about the differential motion vector and information about the reference picture used to predict the current block.

The intra prediction unit 122 predicts pixels within the current block using pixels (reference pixels) located around the current block within the current picture including the current block. There are multiple intra prediction modes depending on the prediction direction, and the surrounding pixels and calculation formulas to be used are defined differently for each prediction mode. In particular, the intra prediction unit 122 can determine the intra prediction mode to be used to encode the current block. In some examples, intra prediction unit 122 may encode the current block using multiple intra prediction modes and select an appropriate intra prediction mode to use from the tested modes. For example, the intra prediction unit 122 calculates rate-distortion values using rate-distortion analysis for several tested intra-prediction modes, and selects the rate-distortion value with the best rate-distortion characteristics among the tested modes. You can also select intra prediction mode.

The intra prediction unit 122 selects one intra prediction mode from a plurality of intra prediction modes and predicts the current block using surrounding pixels (reference pixels) and an operation equation determined according to the selected intra prediction mode. Information about the selected intra prediction mode is encoded by the encoder 150 and transmitted to the video decoding device.

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

The converter 140 converts the residual signal in the residual block with pixel values in the spatial domain into a transform coefficient in the frequency domain. The converter 140 may convert the residual signals in the residual block using the size of the current block as a conversion unit, or divide the residual block into a plurality of smaller subblocks and convert the residual signals into a conversion unit of the subblock size. You can also convert it. There may be various ways to divide the residual block into smaller subblocks. For example, it may be divided into predefined subblocks of the same size, or QT (quadtree) type division may be used with the residual block as the root node.

The quantization unit 145 quantizes the transform coefficients output from the transform unit 140 and outputs the quantized transform coefficients to the encoder 150.

The encoder 150 generates a bitstream by encoding the quantized transform coefficients using an encoding method such as CABAC. In addition, the encoder 150 encodes information such as CTU size, MinQTSize, MaxBTSize, MaxBTDepth, MinBTSize, QT split flag, BT split flag, and split type related to block division, so that the video decoding device is identical to the video encoding device. Allows blocks to be split.

The encoder 150 encodes information about the prediction type indicating whether the current block is encoded by intra prediction or inter prediction, and encodes intra prediction information or inter prediction information depending on the prediction type.

If the current block is intra-predicted, a syntax element for the intra-prediction mode is encoded as intra-prediction information. If the current block is inter predicted, the encoder 150 encodes syntax elements for inter prediction information. Syntax elements for inter prediction information include the following.

(1) Mode information indicating whether the motion information of the current block is encoded in merge mode or in a mode that encodes differential motion vectors.

(2) Syntax elements for motion information

When motion information is encoded by merge mode, the encoder 150 uses merge index information indicating which of the merge candidates is selected as a candidate for extracting motion information of the current block as a syntax element for motion information. Encode.

On the other hand, when the motion information is encoded in a mode that encodes a differential motion vector, information about the differential motion vector and information about the reference picture are encoded as syntax elements for the motion information. If the predicted motion vector is determined by selecting one candidate among a plurality of predicted motion vector candidates, the syntax element for the motion information further includes predicted motion vector identification information to identify the selected candidate. Includes.

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 restores the residual block by converting the transform coefficients output from the inverse quantization unit 160 from the frequency domain to the spatial domain.

The addition unit 170 restores the current block by adding the restored residual block and the prediction block generated by the prediction unit 120. Pixels in the restored current block are used as reference pixels when intra-predicting the next block.

The filter unit 180 performs deblocking filtering on the boundaries between restored blocks to remove blocking artifacts caused by block-level encoding/decoding and stores them in the memory 190. When all blocks in one picture are reconstructed, the reconstructed picture is later used as a reference picture for inter prediction of blocks in the picture to be encoded.

Below, the video decoding device will be described.

Figure 3 is an example block diagram of a video decoding device that can implement the techniques of the present disclosure.

The image decoding device includes a decoding unit 310, an inverse quantization unit 320, an inverse transform unit 330, a prediction unit 340, an adder 350, a filter unit 360, and a memory 370. Like the video encoding device of FIG. 2, each component of the video decoding device may be implemented as a hardware chip, or may be implemented as software and a microprocessor may be implemented to execute the software function corresponding to each component.

The decoder 310 decodes the bitstream received from the video encoding device, extracts information related to block division, determines the current block to be decoded, and provides prediction information and residual signal information necessary to restore the current block. Extract .

The decoder 310 extracts information about the CTU size from the SPS (Sequence Parameter Set) or PPS (Picture Parameter Set), determines the size of the CTU, and divides the picture into CTUs of the determined size. Then, the CTU is determined as the highest layer of the tree structure, that is, the root node, and the CTU is divided using the tree structure by extracting the division information about the CTU. For example, when dividing a CTU using the QTBT structure, first extract the first flag (QT_split_flag) related to the division of the QT and split each node into four nodes of the lower layer. And, for the node corresponding to the leaf node of the QT, the second flag (BT_split_flag) and split type information related to the split of the BT are extracted and the leaf node is split into a BT structure.

Meanwhile, when the decoder 310 determines the current block to be decoded by dividing the tree structure, it extracts information about the prediction type indicating whether the current block is intra-predicted or inter-predicted.

When prediction type information indicates intra prediction, the decoder 310 extracts syntax elements for intra prediction information (intra prediction mode) of the current block.

When the prediction type information indicates inter prediction, the decoder 310 extracts syntax elements for the inter prediction information. First, mode information indicating which of the plurality of encoding modes has been encoded in the motion information of the current block is extracted. Here, the plurality of encoding modes include merge mode and differential motion vector encoding mode. When the mode information indicates a merge mode, the decoder 310 extracts merge index information indicating which of the merge candidates to derive the motion vector of the current block as a syntax element for the motion information. On the other hand, when the mode information indicates the differential motion vector encoding mode, the decoder 310 extracts information about the differential motion vector and information about the reference picture referenced by the motion vector of the current block as syntax elements for the motion vector. do. Meanwhile, when the video encoding device uses one candidate among a plurality of prediction motion vector candidates as the prediction motion vector of the current block, prediction motion vector identification information is included in the bitstream. Therefore, in this case, not only information about the differential motion vector and information about the reference picture, but also prediction motion vector identification information is extracted as a syntax element for the motion vector.

Meanwhile, the decoder 310 extracts information about the quantized transform coefficients of the current block as information about the residual signal.

The inverse quantization unit 320 inversely quantizes the quantized transform coefficients, and the inverse transformation unit 330 inversely transforms the inverse quantized transform coefficients from the frequency domain to the spatial domain to restore the residual signals, thereby generating a residual block for the current block.

The prediction unit 340 includes an intra prediction unit 342 and an inter prediction unit 344. The intra prediction unit 342 is activated when the prediction type of the current block is intra prediction, and the inter prediction unit 344 is activated when the prediction type of the current block is intra prediction.

The intra prediction unit 342 determines the intra prediction mode of the current block among a plurality of intra prediction modes from the syntax elements for the intra prediction mode extracted from the decoder 310, and determines the intra prediction mode of the current block according to the intra prediction mode. Use these to predict the current block.

The inter prediction unit 344 determines motion information of the current block using syntax elements for the intra prediction mode extracted from the decoder 310, and predicts the current block using the determined motion information.

First, the inter prediction unit 344 checks the mode information in the inter prediction extracted from the decoder 310. When the mode information indicates a merge mode, the inter prediction unit 344 constructs a merge list including a predetermined number of merge candidates using neighboring blocks of the current block. The method of the inter prediction unit 344 constructing the merge list is the same as that of the inter prediction unit 124 of the video encoding device. Then, one merge candidate is selected from among the merge candidates in the merge list using the merge index information transmitted from the decoder 310. Then, the motion information of the selected merge candidate, that is, the motion vector and reference picture of the merge candidate, are set as the motion vector and reference picture of the current block.

On the other hand, when the mode information indicates the differential motion vector encoding mode, the inter prediction unit 344 derives prediction motion vector candidates using the motion vectors of neighboring blocks of the current block, and uses the prediction motion vector candidates to predict the current block. Determine the predicted motion vector for the motion vector of . The method by which the inter prediction unit 344 derives predicted motion vector candidates is the same as that of the inter prediction unit 124 of the video encoding device. If the video encoding device uses one candidate among a plurality of predicted motion vector candidates as the predicted motion vector of the current block, the syntax element for the motion information includes predicted motion vector identification information. Therefore, in this case, the inter prediction unit 344 may select the candidate indicated by the prediction motion vector identification information among the prediction motion vector candidates as the prediction motion vector. However, when the video encoding device determines the predicted motion vector using a predefined function for a plurality of predicted motion vector candidates, the inter prediction unit may determine the predicted motion vector by applying the same function as that of the video encoding device. When the predicted motion vector of the current block is determined, the inter prediction unit 344 adds the predicted motion vector and the differential motion vector delivered from the decoder 310 to determine the motion vector of the current block. Then, the reference picture referenced by the motion vector of the current block is determined using information about the reference picture transmitted from the decoder 310.

When the motion vector and reference picture of the current block are determined in merge mode or differential motion vector coding mode, the inter prediction unit 344 generates a prediction block of the current block using the block at the position indicated by the motion vector within the reference picture. do.

Meanwhile, in the case of bi-prediction, the inter prediction unit 344 selects the first reference picture and the second reference picture from reference picture list 0 and reference picture list 1, respectively, using syntax elements for inter prediction information, and each reference A first motion vector and a second motion vector referring to the picture are determined. Then, a first reference block is generated using a first motion vector referring to the first reference picture, and a second reference block is generated using a second motion vector referring to the second reference picture. The prediction block for the current block is generated by averaging or weighted averaging the first reference block and the second reference block.

Additionally, the inter prediction unit 344 may perform the bi-directional optical flow (BIO) process of the present disclosure to generate a prediction block of the current block using bi-directional prediction. In other words, after determining bidirectional motion vectors for the current block, the inter prediction unit 344 can generate a prediction block for the current block using motion compensation according to the BIO process on a pixel-by-pixel or sub-block basis. .

Whether to apply the BIO process when compensating for motion using bidirectional prediction can be determined in various ways. Details on the BIO process and whether to apply the BIO process in the motion compensation process will be described later with reference to the drawings in FIG. 4 and below.

The adder 350 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 restored current block are used as reference pixels when intra-predicting a block to be decoded later.

The filter unit 360 deblocks and filters the boundaries between restored blocks to remove blocking artifacts caused by block-level decoding and stores them in the memory 370. When all blocks in one picture are reconstructed, the reconstructed picture is later used as a reference picture for inter prediction of blocks in the picture to be decoded.

This disclosure relates to the application of bi-directional optical flow (BIO) in motion compensation. The encoding device performs motion estimation and compensation in units of coding blocks (CU: coding units) during the inter-prediction process and then transmits the resulting motion vector (MV) value to the decoding device, and the encoding device and the decoding device Based on the MV value in CU units, the MV value can be additionally corrected on a pixel basis or on a sub-block (i.e., sub-CU) basis that is smaller than the CU using BIO. In other words, BIO can precisely compensate for the movement of coding blocks (CUs) from n×n-sized blocks to 1×1-sized blocks (i.e., pixels) based on the size of each block. Additionally, since the BIO process is performed by applying an explicit equation using previously decoded information shared between the encoding device and the decoding device, signaling of additional information for the BIO process from the encoding device to the decoding device is not required.

Figure 4 is a reference diagram for explaining the basic concept of BIO.

BIO, which is used when encoding and decoding images, is based on the assumption that motion vector information must be bi-prediction information, that the pixels that make up the image move at a constant speed, and that there is little change in pixel values.

First, by (normal) bidirectional motion prediction for the current block, bidirectional motion indicates corresponding regions (i.e., reference blocks) that are most similar to the current block to be encoded in the current picture in the reference pictures (Ref ₀ and Ref ₁ ). Assume that the vectors (MV ₀ and MV ₁ ) have been determined. The two bidirectional motion vectors represent the movement of the current block. In other words, bidirectional motion vectors set the current block as one unit and are an overall estimate of the movement for that unit.

In the example of Figure 4, the pixel in the reference picture (Ref 0) indicated by the bidirectional motion vector (MV ₀ ) corresponding to the pixel P in the current block is P 0, and the pixel in the reference picture (Ref ₀ ) indicated by the bidirectional motion vector (MV 0) is P ₀ , and the bidirectional motion vector (MV 0) corresponding to the pixel P in the current block is The pixel in the reference picture (Ref ₁ ) pointed to by MV ₁ ) is P ₁ . And, let us assume that the movement of pixel P in FIG. 4 is slightly different from the overall movement of the current block. For example, if the object located at pixel A in Ref ₀ in FIG. 4 moves through pixel P in the current block of the current picture and moves to pixel B in Ref1, pixel A and pixel B have quite similar values. Also, in this case, the point in Ref0 that is most similar to pixel P in the current block is not P0 pointed to by the bidirectional motion vector (MV0), but A that moves P0 by a predetermined displacement vector ( v _x τ ₀ , v _y τ ₀ ). , and the point in Ref1 that is most similar to pixel P in the current block is not P1 indicated by the bidirectional motion vector (MV1), but B, where P1 is moved by a predetermined displacement vector (- v _x τ ₁ , -v _y τ ₁ ). . τ ₀ and τ ₁ refer to the time axis distance to Ref0 and Ref1, respectively, based on the current picture, and are calculated based on POC (Picture Order Count). Hereinafter, for convenience, ( v _x , v _y ) is referred to as “optical flow” or “BIO motion vector.”

Therefore, when predicting the pixel (P) value of the current block in the current picture, using the values of the two reference pixels (A, B), the reference pixels (P0, P1) indicated by the bidirectional motion vectors (MV0, MV1) More accurate predictions are possible than using As above, the concept of changing the reference pixel used to predict one pixel of the current block by considering _the movement _of pixels within the current block specified by the optical flow ( v It can be expanded to consider the movement of units.

Below, we will describe a theoretical method for generating predicted values for pixels in the current block according to the BIO technique. For convenience of explanation, it is assumed that BIO-based bidirectional motion compensation is performed on a pixel basis.

A bidirectional motion vector that points to corresponding regions (i.e., reference blocks) most similar to the current block to be coded in the reference picture (Ref ₀ and Ref ₁ ) by (normal) block-wise bidirectional motion prediction for the current block. Assume that (MV ₀ ,MV ₁ ) has been determined. The decoding device can determine bidirectional motion vectors (MV ₀ and MV ₁ ) from motion vector information included in the bitstream. In addition, the luminance value of the pixel in the reference picture (Ref ₀ ) indicated by the motion vectors (MV ₀ ) corresponding to the pixel ( i,j ) in the current block is called I ⁽⁰⁾ ( i,j ). Define the brightness value of the pixel in the reference picture (Ref ₁ ) pointed to by the motion vectors (MV ₁ ), which corresponds to the pixel ( i,j ) in the current block, as I ⁽¹⁾ ( i,j ).

The brightness value of pixel A in the reference picture Ref _{0 ,} which the _BIO motion vector ( v _x , v _y ) indicates corresponds to the pixel in the current block, _is I ( ₀ ⁾ ( i + v It can be defined as, and the brightness value of pixel B in the reference picture Ref ₁ can be defined as I ⁽¹⁾ ( i - v _x τ ₁ , j - v _y τ ₁ ). Here, if linear approximation is performed using only the first order term of the Taylor series, A and B can be expressed as Equation 1.

Here, I _x ^{( k )} and I _y ^{( k )} ( k = 0, 1) represent the gradient values in the horizontal and vertical directions at the ( i , j ) positions of Ref0 and Ref1. τ ₀ and τ ₁ refer to the time axis distance to Ref0 and Ref1, respectively, based on the current picture, and are calculated based on POC (Picture Order Count). The formula is τ ₀ = POC(current) - POC(Ref0), τ ₁ = POC(Ref1) - POC(current).

The bidirectional optical flow ( v _x , v _y ) of each pixel in the block is determined as a solution that minimizes Δ, which is defined as the difference between pixel A and pixel B. Using the linear approximation of A and B derived in Equation 1, Δ can be defined as Equation 2.

For brevity, the pixel location ( i , j ) is omitted in each term of Equation 2 above.

For more robust optical flow estimation, we introduce the assumption that motion is locally consistent with surrounding pixels. The BIO motion vector for the pixel ( i , j ) to be currently predicted is all pixels present in the mask Ω of size (2M+1)x(2M+1) centered on the pixel ( i , j ) to be currently predicted. Consider the difference values Δ in Equation 2 for ( i' , j' ). In other words, the optical flow for the current pixel ₍ i , j ) is the objective function Φ( v , v _y ) can be determined as the vector that minimizes it.

Here, ( i' , j' ) means the positions of pixels within the mask Ω. For example, when M = 2, the shape of the mask is as shown in FIG. 5. The hatched pixel located in the center of the mask is the current pixel ( i , j ), and the pixels within the mask Ω are expressed as ( i' , j' ).

To obtain the optical flow ( v _x , v _y ) of each pixel ( i , j ) in the block, the solution that minimizes the objective function Φ( v _x , v _y ) is calculated using an analytic method. Partially differentiate _the objective function _Φ ( v _x , v _{y )} with v _x and v _y , _respectively , to get _∂Φ ( _v By deriving 0 and combining the two equations, Equation 4 can be obtained.

s ₁ , s ₂ , s ₃ , s ₄ , s ₅ , and s ₆ in Equation 4 are the same as Equation 5.

Here, since s ₂ = s ₄ , s ₄ is replaced with s ₂ .

You can find v _x and v _y by solving the simultaneous equations in Equation 4. For example, deriving v _x and v _y using Cramer's rule is as shown in Equation 6.

As another example, a simplified method of calculating the approximate value of v _x by substituting v _y = 0 into the first equation of Equation 4, and calculating the approximate value of v y by substituting v _x calculated _first into the second equation. can also be used, in which case v _x and v _y are expressed as Equation 7.

Here, r and m are regularization parameters introduced to avoid division being performed by 0 or very low values. In equation (7), if s ₁ + r > m is not satisfied, v _x ( i , j ) = 0 is set, and if s ₅ + r > m is not satisfied, Set v _y ( i , j ) = 0.

As another example, you can use the method of calculating the approximate value of v _x by substituting v _y = 0 into the first equation of Equation 4, and calculating the approximate value of v _y by substituting v _x = 0 into the second equation. there is. Using this method, v _x and v _y can be calculated independently, and v _x and v _y are expressed as Equation 8.

_{As another} example , _calculate the _{approximate value} of v You can use the method of calculating v _y as the average value of the second approximation of v _y obtained by substituting 0. If v _x and v _y are obtained using this method, it is equivalent to Equation 9.

The normalization parameters r and m used in Equations 7 to 9 can be defined as Equation 10.

Here, d means the bit depth of the pixels of the image.

The optical flow v _x and v _y of each pixel in the block is obtained by calculating for each pixel in the block using methods such as Equation 6 to Equation 9.

When _the optical _flow ( v _ _{_}

In Equation 11, ( I ⁽⁰⁾ + I ⁽¹⁾ )/2 is a typical block-wise bidirectional motion compensation, so the remaining terms can be referred to as BIO offset.

Meanwhile, in typical two-way motion compensation, a prediction block of the current block is generated using pixels in the reference block. On the other hand, the use of a mask must allow access to pixels other than those in the reference block. For example, the mask for the pixel at the upper leftmost position ((0,0) position) of the reference block as shown in FIG. 6(a) includes pixels located at positions outside the reference block. To maintain the same memory access as in conventional bidirectional motion compensation and to reduce the computational complexity of BIO, the I ^(k) , I _x ^(k) , I _y ^(k) of pixels outside the reference block within the mask are referenced. It can be used by padding with the corresponding values of the nearest pixel within the block. For example, as shown in Figure 6(b), when the size of the mask is 5x5, I ^(k) , I _x ^(k) , and I _y ^(k) for external pixels located at the top of the reference block are ^The pixels of the _{top row} within can ^be padded with I ^{( k )} _, I ⁾ ,I _y ^(k) can also be padded with I ^(k) ,I _x ^(k) ,I _y ^(k) in the leftmost column.

In the above, the BIO process performed for each pixel in the current block has been described. However, to reduce computational complexity, the BIO process may be performed on a block basis, for example, in 4x4 units. BIO performed in units of subblocks within the current block can obtain optical flows v _x and v _y in units of each subblock within the current block using methods such as equations (6) to (9). Subblock-based BIO has the same principle as pixel-level BIO, only the mask range is different.

As an example, the range of the mask Ω may be expanded to include the range of subblocks. When the size of the subblock is N×N, the mask Ω has a size of (2M+N)×(2M+N). For example, when M = 2 and the size of the subblock is 4 × 4, the mask has the shape shown in FIG. 7. Calculate Δ in Equation 2 for all pixels in the mask Ω including the subblock to obtain the objective function in Equation 3 for the subblock _, and apply Equations 4 to 9 to calculate the optical flow ( v ,v _y ) can be calculated.

As another example, the objective function of Equation 3 for the subblock can be obtained by calculating the Δs in Equation 2 by applying a mask to all pixels in the subblock on a per-pixel basis and by summing the squares of the calculated Δs. Then, the optical flow ( v _x , v _y ) for the subblock can be calculated by minimizing the objective function. For example, referring to FIG. 8, a 5x5 sized mask 810a is applied to the pixel at the (0,0) position of the 4x4 subblock 820 in the current block, and Equation 2 is applied to all pixels in the mask 810a. Calculate the Δ values of . Then, a 5x5 sized mask 810b is applied to the pixel at the (0,1) position, and Δ values in Equation 2 are calculated for all pixels within the mask 810b. Through this process, the objective function of Equation 3 can be obtained by summing the squares of Δ calculated for all pixels in the subblock. And we can calculate the optical flow ( v _x , v _y ) that minimizes the objective function. The objective function in this example is expressed as Equation 12.

Here, b _k represents the k-th subblock in the current block, and Ω(x,y) represents a mask for a pixel with (x, y) coordinates in the k-th subblock. And, s ₁ to s ₆ used for optical flow ( v _x , v _y ) calculation are modified as shown in Equation 13.

In the above equation and represents I _x ^(k) and I _y ^(k) , that is, the horizontal gradient and the vertical gradient, respectively.

As another example, a method of using a mask in subblock units as shown in FIG. 7 and applying weights to each position of the mask may be used. Higher weights are applied toward the center of the subblock. For example, referring to FIG. 8, when a mask is applied to each pixel within a subblock, Δ's for the same position may be calculated twice. Most of the pixels located in the mask 810a centered on the pixel at the (0,0) position of the subblock 820 are masked ( It is also located within 810b). Therefore, Δs may be calculated redundantly. Instead of repeatedly calculating overlapping Δ, each position in the mask can be given a weight based on the number of overlapping times. For example, if M = 2 and the size of the subblock is 4 × 4, it may have a weighted mask form as shown in FIG. 9. By taking this approach, the computation complexity of Equations 12 and 13 can be simplified and computational complexity can be reduced.

The pixel-based or subblock-based BIO described above requires a large amount of computation. Therefore, a method for reducing the amount of computation due to BIO in video encoding or decoding is required, and this disclosure proposes skipping the BIO process in motion compensation when certain conditions are met to reduce the amount of computation.

Figure 10 is a block diagram showing the configuration of a device that performs motion compensation by selectively applying a BIO process according to an embodiment of the present invention.

The motion compensation device 1000 described in this embodiment may be implemented in the inter prediction unit 124 of an image encoding device and/or the inter prediction unit 344 of an image decoding device, and may include a reference block generator 1010, It may include a skip decision unit 1020 and a prediction block generator 1030. Each of these components may be implemented as a hardware chip, or may be implemented as software and one or more microprocessors may be implemented to execute the functions of the software corresponding to each component.

The reference block generator 1010 generates a first reference block using a first motion vector referring to the first reference picture in reference picture list 0, and a second reference block referring to the second reference picture in reference picture list 1. A second reference block is created using the motion vector.

The skip decision unit 1020 determines whether to apply the BIO process in the motion compensation process.

When the skip decision unit 1020 determines that the BIO process is skipped, the prediction block generator 1030 generates a prediction block of the current block using typical motion compensation. That is, the prediction block of the current block is generated by averaging or weighting the first reference block and the second reference block. On the other hand, if it is determined by the skip decision unit 1020 that the BIO process is applied, the prediction block generator 1030 generates a prediction block of the current block using the first reference block and the second reference block according to the BIO process. do. In other words, a prediction block of the current block can be generated by applying Equation 11.

The skip decision unit 1020 may determine whether to apply the BIO process based on one or more of the following conditions.

- Texture complexity of the current block

- Mode information indicating the size of the current block and/or motion information encoding mode

- Whether the bidirectional motion vector (first motion vector and second motion vector) satisfies the constant velocity constraint (CVC) and/or brightness constancy constraint (BCC)

- Degree of variation in motion vectors of surrounding blocks

Below, a specific method for determining whether to apply the BIO process using each condition will be described.

Example 1: BIO skip according to texture complexity

Optical flow tends to show unreliable results in flat areas with few local features, such as edges or corners. In addition, it is highly likely that areas with such a flat texture have already been sufficiently well predicted through conventional block-level motion estimation. Therefore, in this embodiment, the texture complexity of the current block is calculated and the BIO process is skipped according to the texture complexity.

In order to allow the encoding device and the decoding device to calculate the texture complexity on their own without additional signaling, the texture complexity of the current block can be calculated using the first and second reference blocks shared by the encoding device and the decoding device. there is. In other words, the skip decision unit implemented in each of the encoding device and the decoding device calculates the texture complexity of the current block and determines whether to skip the BIO.

Meanwhile, for texture complexity, local feature detectors with a small amount of calculation, such as difference values from surrounding pixels, gradients, and Moravec, can be used. In this embodiment, texture complexity is calculated using gradient. Since the gradient for reference blocks is a value used in the BIO process, using the gradient has the advantage of being able to apply the gradient values calculated from the texture complexity as is when performing the BIO process.

The motion compensation device 1000 of this embodiment calculates texture complexity using the horizontal gradient and vertical gradient of each pixel in the first and second reference blocks. As an example, the motion compensation device 1000 calculates the horizontal complexity using the horizontal gradient of each pixel in the first reference block and the second reference block, and calculates the horizontal complexity of the first reference block and the second reference block. The vertical complexity is calculated using the vertical gradient of each pixel in the pixel. For example, horizontal complexity and vertical complexity can be calculated by Equation 14.

Here, D1 and D5 represent the horizontal complexity and vertical complexity, respectively, and CU is a set of pixel positions in the first reference block and the second reference block corresponding to the position of each pixel in the current block, [i, j]. represents the position corresponding to each pixel in the current block in the first and second reference blocks. And, d ₁ (i,j) and d ₅ (i,j) can be calculated by Equation 15.

The horizontal complexity and vertical complexity of Equation 14 can be calculated using d ₁ and d ₅ in Equation 15. _That is , _the ^horizontal gradients ₍ τ ₀ I The _horizontal complexity _{D 1} ^can be calculated by calculating the sum of ,τ ₁ I And, vertical gradients ( τ ₀ I _y ⁽⁰⁾ (i,j), τ ₁ I _y ^{(1 )} You can calculate the vertical complexity D ₅ by calculating the sum of (i,j) ) for all pixel positions and summing their squares.

d ₄ is missing in Equation 15, but d ₄ has the same value as d ₂ . It can be seen that d ₁ to d ₆ in Equation 15 are related to s ₁ to s ₆ in Equation 5. d ₁ to d ₆ mean values at one pixel position, and s ₁ to s ₆ mean the sum of d ₁ to d ₆ calculated at all pixel positions in a mask centered on one pixel. In other words, using Equation 15, Equation 5 can be expressed as the following Equation 16. In Equation 16, s ₄ has the same value as s ₂ , so its description is omitted.

The texture complexity for the current block is the minimum value (Min( D ₁ , D ₅ )), maximum value (Max( D ₁ , D ₅ )), or average value (Ave( D ₁ , D ₅ )) between the horizontal and vertical complexity. ) can be set to any one of the following. The motion compensation device 1000 skips the BIO process if the texture complexity is less than the threshold T, and applies the BIO process if the texture complexity is greater than or equal to the threshold T. When applying the BIO process, d ₁ to d ₆ calculated in Equation 14 can be used as is for the calculation of s ₁ to s ₆ . In other words, this embodiment obtains the texture complexity of the current block using the value that must be calculated during the BIO process and uses this to determine whether to skip BIO, so it has the advantage of reducing the amount of additional calculation to determine whether to skip BIO.

Meanwhile, the threshold T can be determined by scaling the normalization parameters used in Equations 7 to 9. The normalization parameters r and m have the relationships s ₁ > m - r and s ₅ > m - r , and if s ₁ <= m - r , v _x becomes 0 even if BIO is performed, and s ₅ <= m - In case of r , v _y becomes 0 even if BIO is performed.

Therefore, if you set the threshold T based on the relational expression of the normalization parameter, even if BIO is performed, the area that becomes 0 can be determined in advance on a CU basis and the BIO can be skipped. D ₁ is the sum of d ₁ for all pixel positions in the CU, and s ₁ is the sum of all d ₁ in the mask Ω, so the size of the CU is W×H and the size of the mask Ω is (2M+1)×( When 2M+1), the threshold T can be set as in Equation 17.

Figure 11 is an example diagram of a process for performing motion compensation by selectively applying a BIO process according to the texture complexity of the current block according to this embodiment.

The motion compensation device 1000 calculates the horizontal gradient (I _x ^(k) ) and the vertical gradient (I _y ^(k) ) for each pixel in the first and second reference blocks (S1102). Then, d ₁ to d ₆ are calculated using Equation 15, and among them, d ₁ and d ₅ are used to calculate the horizontal complexity (D ₁ ) and vertical complexity (D ₅ ) according to Equation 14. (S1104). It is determined whether the texture complexity of the current block, which is the minimum of the horizontal complexity (D ₁ ) and the vertical complexity (D ₅ ), is less than the threshold (T) (S1106). In this example, the texture complexity of the current block is explained as being the minimum of the horizontal complexity (D ₁ ) and the vertical complexity (D ₅ ), but as described above, the texture complexity may be set to the maximum or average value.

If the texture complexity of the current block is less than the threshold (T), the BIO process is skipped and a prediction block of the current block is generated by normal motion compensation (S1108). That is, the prediction block of the current block is generated by averaging or weighting the first reference block and the second reference block.

On the other hand, if the texture complexity of the current block is greater than or equal to the threshold (T), a prediction block for the current block is generated using the first and second reference blocks according to the BIO process. First, calculate s ₁ to s ₆ . At this time, since the horizontal and vertical gradients for pixels in the reference blocks have already been calculated in S1102, in order to obtain s ₁ to s ₆ , the horizontal and vertical gradients are calculated only for pixels outside the reference block present in the mask. Just calculate the gradients. Or, as described above, when the horizontal gradient and vertical gradient for pixels outside the reference block are padded with the corresponding values of nearby pixels inside the reference block, the horizontal gradient for pixels within the already calculated reference blocks s ₁ to s ₆ can also be obtained using only the direction and vertical gradients.

Alternatively, since d ₁ to d ₆ are related to s ₁ to s ₆ (see Equation 16), the already calculated d ₁ to d ₆ may be used to calculate s ₁ to s ₆ .

When s ₁ to s ₆ are calculated, the optical flow ( v _x , v _y ) in pixel or subblock units is determined using any one of Equations 6 to 9 (S1112). _Then , optical flow ₍ v

Figure 12 is another example diagram showing the process of performing motion compensation by selectively applying the BIO process according to the texture complexity of the current block according to this embodiment.

The example disclosed in FIG. 12 is different from FIG. 11 only in the order of calculating d ₁ to d ₆ . In other words, in order to calculate the texture complexity of the current block, only d ₁ and d ₅ among d ₁ to d ₆ are needed. Therefore, as in S1204, d ₁ and d ₅ are first obtained to calculate texture complexity, and the remaining d2, d3, d4 (same as d2), and d5 are calculated when the texture complexity is greater than the threshold and the BIO process is performed. (S1210). Other processes are substantially the same as in FIG. 11.

The table below shows the results of an experiment comparing motion compensation based on the BIO process and motion compensation using the BIO process selectively based on texture complexity according to this embodiment.

The sequences used in the experiment were 4 Class A1 (4K), 5 Class B (FHD), 4 Class C (832×480), and 4 Class D (416×240), and the entire frame of each video was tested. did. The experimental environment was Random Access (RA) configuration, and the QP values were set to 22, 27, 32, and 37 to compare BD rates.

According to this example, on average, about 19% of BIO was skipped, and in Class A1 (4K), which has the largest computation burden, 32% of BIO was skipped. It can be seen that the skipping rate tends to increase as the resolution of the image increases, and in reality, the higher the resolution, the greater the burden of computation, so this can be considered a meaningful result.

Additionally, there was an average increase in the Y BD rate of 0.02%, but a BD rate difference of less than 0.1% is generally accepted as a negligible difference. Therefore, according to this example, it can be seen that even if BIO is selectively skipped, the compression efficiency is almost the same.

Meanwhile, the examples described above are related to deciding whether to skip the entire BIO process. However, instead of skipping the entire BIO process, it is also possible to independently skip the horizontal optical flow ( v _x ) and vertical optical flow ( v _y ). In other words, if the complexity (D ₁ ) in the horizontal direction is less than the threshold (T), v _x = 0 and the BIO process for the horizontal direction is skipped, and the complexity (D ₅ ) in the vertical direction is less than the threshold (T). If it is small, set v _y = 0 and skip the BIO process for the vertical direction.

Figure 13 is another example of the process of performing motion compensation by selectively applying the BIO process according to the texture complexity of the current block according to this embodiment.

The motion compensation device 1000 calculates the horizontal gradient (I _x ^(k) ) and the vertical gradient (I _y ^(k) ) for each pixel in the first and second reference blocks (S1310). Then, calculate d ₁ and d ₅ using Equation 15, calculate the horizontal complexity (D ₁ ) using d ₁ , and calculate the vertical complexity (D ₅ ) using d ₅ (S1320).

When the horizontal complexity (D ₁ ) and the vertical complexity (D ₅ ) are calculated in S1320, a process of determining whether to skip the horizontal optical flow (S1330) and whether to skip the vertical optical flow are performed. Perform the decision process (S1340). In Figure 13, it is explained that the horizontal optical flow skip is determined first, but it is of course also possible to determine the vertical optical flow skip first.

First, in S1330, the motion compensation device 1000 determines whether the horizontal complexity (D ₁ ) is less than the threshold (T) (S1331). If the horizontal complexity (D ₁ ) is less than the threshold (T), the horizontal optical flow ( v _x ) is set to 0 (S1332). This means that horizontal optical flow is not applied. If the horizontal complexity (D ₁ ) is greater than or equal to the threshold (T), d ₃ is calculated (S1333), and s ₁ and s ₃ are calculated using d ₁ and d ₃ (S1334). Referring to Equations 7 to 9, when calculating the horizontal optical flow ( v _x ), only s ₁ and s ₃ are required. Since d ₁ has already been calculated in S1320, d ₃ is calculated in S1333, and s ₁ and s ₃ are calculated using d ₁ and d ₃ in S1334. Then, the horizontal optical flow ( v _x ) is calculated using s ₁ and s ₃ according to any one of Equations 7 to 9 (S1335).

Afterwards, the process proceeds to S1340, which determines whether to skip the vertical optical flow. Determine whether the vertical complexity (D ₅ ) is less than the threshold (T) (S1341). If the vertical complexity (D ₅ ) is less than the threshold (T), the vertical optical flow ( v _y ) is set to 0 (S1342). This means that vertical optical flow is not applied. If the vertical complexity (D ₅ ) is greater than or equal to the threshold (T), calculate d ₂ and d ₆ (S1343), and use d ₂ , d ₅ , and d ₆ to calculate s ₂ , s ₅ , and s ₆ Calculate (S1344). When calculating the vertical optical flow ( v _y ) using Equation 7 or 9, only s ₂ , s ₅ , and s ₆ are required. Since d ₅ has already been calculated in S1320, d ₂ and d ₆ are calculated in S1343, and s ₂ , s ₅ , and s ₆ are calculated using d ₂ , d ₅ , and d ₆ in S1344. Then, calculate the vertical optical flow ( v _y ) using s ₂ , s ₅ , and s ₆ according to either Equation 7 or 9 (S1345).

Meanwhile, if calculating the vertical optical flow ( v _y ) using Equation 8, only s ₅ and s ₆ are required. Therefore, in this case, the calculation of d ₂ and s ₂ may be omitted in S1343 and S1344.

By substituting the horizontal optical flow ( v _x ) and vertical optical flow ( v _y ) calculated in this way into Equation 11, a prediction block of the current block is generated. In Equation 11, when the horizontal optical flow is skipped, v _x = 0, so the horizontal optical flow ( v _x ) does not contribute to prediction block generation. Similarly, when the vertical optical flow is skipped, v _y = 0, so the vertical optical flow ( v _x ) does not contribute to prediction block generation. If the horizontal and vertical optical flows are mode-skipped, v _x = 0 and v _y = 0, the prediction block is generated by averaging the first and second reference blocks. In other words, a prediction block is generated through normal motion compensation.

In Example 1 described above, the texture complexity of the current block is estimated using pixels in the reference block. However, the texture complexity of the current block can also be calculated using actual pixels within the current block. For example, the encoding device may calculate the horizontal complexity and vertical complexity using the horizontal and vertical gradients of pixels in the current block. That is, the horizontal complexity is calculated using the sum of squares of the horizontal gradients of each pixel in the current block, and the vertical complexity is calculated using the sum of squares of the vertical gradients. Then, the horizontal and vertical complexities are used to decide whether to skip the BIO process. In this case, unlike the encoding device, the decoding device does not know the pixels in the current block, so it cannot calculate texture complexity in the same way as the encoding device. Therefore, the encoding device must additionally signal information indicating whether to skip BIO to the decoding device. That is, the skip decision unit implemented in the decoding device decodes the information indicating whether to skip the BIO received from the encoding device and selectively skips the BIO process according to what the information indicates.

Example 2: BIO skip according to the size of the current block and/or motion information encoding mode

As described above, depending on the division of the tree structure from the CTU, the CU corresponding to the leaf node of the tree structure, that is, the current block, may have various sizes.

If the size of the current block is sufficiently small, the motion vector of the current block is likely to have a value quite similar to the BIO in pixel or subblock units, so the correction effect obtained by performing BIO may be small. In this case, the reduction in complexity achieved by skipping BIO is likely to be a greater benefit than the loss of precision due to skipping BIO.

Meanwhile, the motion vector of the current block may be encoded in a merge mode or a mode that encodes a differential motion vector, as described above. If the motion vector of the current block is encoded in merge mode, the motion vector of the current block is merged with the motion vectors of the surrounding blocks. In other words, the motion vector of the current block is set to be the same as the motion vector of the surrounding blocks. In this case, additional correction effects can be obtained through BIO.

Therefore, this embodiment skips the BIO process using at least one of mode information indicating the size of the current block or the encoding mode of the motion vector.

Figure 14 is an example diagram illustrating a process of performing motion compensation by selectively applying a BIO process according to the size of the current block and the encoding mode of the motion vector according to this embodiment. In Figure 14, whether to skip BIO is determined using both the size of the current block and the encoding mode of the motion vector, but using either one is also within the scope of the present invention.

The motion compensation device 1000 first determines whether the size of the current block (CU), which is the encoding target block, is less than or equal to the threshold size (S1402). If the size of the current block (CU) is larger than the threshold size, a prediction block of the current block is generated according to the BIO process (S1408).

On the other hand, if the size of the current block (CU) is less than the threshold size, it is determined whether the motion vector (MV) of the current block (CU) has been encoded by merge mode (S1404). If it is not encoded by merge mode, the BIO process is skipped and a prediction block of the current block is generated through normal motion compensation (S1406). If encoded by merge mode, a prediction block of the current block is generated according to the BIO process (S1408).

For example, when _{w t} If not encoded, the BIO process is skipped.

Meanwhile, in the process of generating a prediction block according to the BIO process in S1308, it is possible to further determine whether to skip BIO according to Example 1, that is, according to the texture complexity of the current block.

Example 3: CVC and/or to BCC BIO according to skip

BIO is based on the assumption that objects in an image move at a constant speed and that pixel values rarely change. These are defined as constant velocity constraint (CVC) and brightness constancy constraint (BCC), respectively.

_{If the} bidirectional motion vectors ₍ MV _{_} It is likely to have a value similar to the bidirectional motion vector of the current block.

_The fact that _the current block 's _{bidirectional} motion _vectors ( MV .

_{That the} bidirectional motion vector of the current block satisfies the BCC condition means that the first reference block _located in the first reference picture ( Ref0 ) indicated by ₍ MV 2 This means that the difference between the reference blocks located in the reference picture (Ref1) is 0. The difference between two reference blocks can be calculated by Sum of Absolute Difference (SAD), Sum of Squared Errors (SSE), etc.

As an example, the CVC condition and BCC condition can be expressed as follows.

Here, T _CVC and T _BCC are the threshold values of the CVC condition and BCC condition, respectively.

Referring to FIG. 4, BIO has optical flow (+ v _x , + v _y ) for the first reference picture (Ref0) and optical flow (- v _x , - v _y ) for the second reference picture (Ref1). We start from the assumption that sizes have different signs. Therefore , bidirectional motion vectors (MV x _{0 ,} MV y ₀ ) and (MV x ₁ ,MV y ₁ ) satisfy the assumption _of BIO, which means that the This means that they must have different signs, and the y components (MV y ₀ and MV y ₁ ) must also have different signs. In addition, _in order to satisfy the constant velocity condition _, the absolute value _of MV The absolute magnitude of the value divided by τ ₁ , the time axis distance, must be the same. Likewise, the absolute magnitude of MV y ₀ divided by τ ₀ and MV y ₁ divided by τ ₁ must be the same. Therefore, if the concept of threshold is introduced, it can be expressed as the CVC condition above.

Meanwhile, the BCC condition is satisfied when the SAD between reference blocks referenced by the bidirectional motion vectors (MV x ₀ , MV y ₀ ) and (MV x ₁ , MV y ₁ ) is less than or equal to the threshold (T _BCC ). Of course, other indicators that can represent the difference between two reference blocks, such as SSE, may be used instead of SAD.

Figure 15 is an example diagram showing a process of performing motion compensation by selectively applying a BIO process according to the CVC condition and BCC condition according to this embodiment.

The motion compensation device 1000 determines whether the bidirectional motion vectors (MV x ₀ , MV y ₀ ) and (MV x ₁ , MV y ₁ ) of the current block satisfy the CVC condition and BCC condition (S1502). If all conditions are met, the BIO process is skipped and a prediction block is generated according to normal motion compensation (S1504).

On the other hand, if either of the two conditions is not met, a prediction block of the current block is generated according to the BIO process (S1506).

In Figure 15, it is explained that the BIO process is skipped when both the CVC condition and the BCC condition are met, but this is only an example and it is okay to decide whether to skip the BIO using either the CVC condition or the BCC condition. do.

Example 4: BIO skip according to the degree of variation of motion vectors of neighboring blocks

If the bidirectional motion vectors estimated on a block basis in neighboring blocks of the current block have similar values, the optical flow estimated on a pixel or subblock basis within the current block is also likely to have similar values.

Therefore, it is possible to determine whether to skip the BIO of the current block based on the degree of variation of the motion vectors of neighboring blocks, for example, variance or standard deviation. As an extreme example, if the motion vector variance of neighboring blocks is 0, the optical flow in units of pixels or subblocks within the current block is also likely to have the same value as the motion vector of the current block, so BIO is skipped.

As an example, the motion vector distribution of neighboring blocks can be expressed as Equation 18.

Here, L is the set of neighboring blocks and l is the total number of neighboring blocks. ( m , n ) represents the index of the surrounding block and t ∈ (0, 1).

Figure 16 is an example diagram illustrating a process of performing motion compensation by selectively applying a BIO process according to the motion vector variance values of neighboring blocks according to this embodiment.

The motion compensation device 1000 compares the variance of motion vectors of neighboring blocks with a predetermined threshold (S1602). If the motion vector variance of the neighboring block is lower than the threshold, the BIO process is skipped and a prediction block is generated according to normal motion compensation (S1604). On the other hand, if the motion vector variance of the neighboring block is higher than the threshold, a prediction block of the current block is generated according to the BIO process (S1606).

In the above, it was explained through Examples 1 to 4 that each condition was used individually to determine whether to skip BIO. However, the present invention is not limited to determining whether to skip BIO using any one condition, and determining whether to skip BIO by selectively combining a plurality of conditions described in the present disclosure is also included in the scope of the present invention. It should be interpreted as For example, a method of determining whether to skip BIO using the size condition of the current block and the texture complexity of the current block, a method of determining whether to skip BIO using the size condition of the current block and the CVC condition and/or BCC condition, and the CVC condition. and a method of determining whether to skip BIO using one or more of the BCC conditions and the texture complexity of the current block, etc., should be understood as being included in the scope of the present invention.

The above description is merely an illustrative explanation of the technical idea of the present embodiment, and those skilled in the art will be able to make various modifications and variations without departing from the essential characteristics of the present embodiment. Accordingly, the present embodiments are not intended to limit the technical idea of the present embodiment, but rather to explain it, and the scope of the technical idea of the present embodiment is not limited by these examples. The scope of protection of this embodiment should be interpreted in accordance with the claims below, and all technical ideas within the equivalent scope should be interpreted as being included in the scope of rights of this embodiment.

Claims (12)

In the inter prediction device that predicts the current block using bi-directional optical flow in video encoding,
means for generating a first motion vector for a first reference picture and a second motion vector for a second reference picture;
To determine whether to apply the bidirectional optical flow from difference values between samples in the first reference picture determined based on the first motion vector and samples in the second reference picture determined based on the second motion vector. A means of deriving the variables used; and
Depending on the variable, prediction samples for the current block are selectively generated from the first reference picture and the second reference picture by applying the bidirectional optical flow, or the first reference is generated without applying the bidirectional optical flow. Means for generating prediction samples for the current block from a picture and the second reference picture
Inter prediction device comprising: ◈Claim 2 was abandoned upon payment of the setup registration fee.◈ According to paragraph 1,
The variable derived from the difference values is SAD (Sum of Absolute Difference) or SSE (Sum of Squared Errors). According to paragraph 1,
If the variable is less than a predetermined threshold, the bidirectional optical flow is not applied,
Inter prediction device, characterized in that applying the bidirectional optical flow when the singi variable is greater than the threshold. According to paragraph 1,
If either the width or height of the current block is smaller than a predefined length, the bidirectional optical flow is not applied. According to paragraph 4,
Inter prediction device, characterized in that the predefined length is 8. In the inter prediction method of predicting the current block using bi-directional optical flow in video decoding,
Generating a first motion vector for a first reference picture and a second motion vector for a second reference picture;
To determine whether to apply the bidirectional optical flow from difference values between samples in the first reference picture determined based on the first motion vector and samples in the second reference picture determined based on the second motion vector. Deriving variables to be used; and
Including generating prediction samples for the current block,
The prediction samples are selectively generated from the first reference picture and the second reference picture by applying the bidirectional optical flow, or are generated from the first reference picture and the second reference picture without applying the bidirectional optical flow, depending on the variable. An inter prediction method characterized by being generated from a second reference picture. ◈Claim 7 was abandoned upon payment of the setup registration fee.◈ According to clause 6,
The inter prediction method, characterized in that the variable derived from the difference values is SAD (Sum of Absolute Difference) or SSE (Sum of Squared Errors). According to clause 6,
The step of generating the prediction samples is,
generating the prediction samples without applying the bidirectional optical flow if the variable is smaller than a predetermined threshold; and
If the singi variable is greater than the threshold, generating the prediction samples by applying the bidirectional optical flow.
An inter prediction method comprising: According to clause 6,
An inter prediction method characterized in that if either the width or height of the current block is smaller than a predefined length, the bidirectional optical flow is not applied. According to clause 9,
Inter prediction method, characterized in that the predefined length is 8. A recording medium readable by a decoder that stores a bitstream generated by an inter prediction method that predicts the current block using bi-directional optical flow,
The inter prediction method is,
Generating a first motion vector for a first reference picture and a second motion vector for a second reference picture;
To determine whether to apply the bidirectional optical flow from difference values between samples in the first reference picture determined based on the first motion vector and samples in the second reference picture determined based on the second motion vector. Deriving variables to be used; and
Including generating prediction samples for the current block,
The prediction samples are selectively generated from the first reference picture and the second reference picture by applying the bidirectional optical flow, or are generated from the first reference picture and the second reference picture without applying the bidirectional optical flow, depending on the variable. A recording medium, characterized in that it is generated from a second reference picture. delete

Images