CN104539970A - 3D-HEVC interframe coding merge mode fast decision making method - Google Patents

Abstract

The present invention discloses a fast decision-making method for 3D-HEVC inter-frame coding merging mode. 1) Extract a CU to be coded from a texture map of a dependent viewpoint. 2) The current CU checks merge mode and merge-skip mode in 2N×2N partition mode. 3) Determine whether the frame where the current CU is located is a P frame or a B frame. 4) Through the judgment of the previous step, if the obtained result is a P frame, judge whether the division level of the current CU is 3 and whether the condition a and condition b are satisfied at the same time; 5) Through the judgment of step 3, if the obtained result is a B frame, The 5 reference blocks of the current CU in the view direction are determined by the disparity vector (DoNBDV) obtained from the adjacent blocks, including the corresponding block of the current CU in the view direction and its 4 adjacent blocks. Judging whether the current CU satisfies condition c and condition d at the same time: the method of the present invention can save an average of about 6.5% of total encoding time and 20.4% of non-independent view texture map encoding time while ensuring encoding quality.

Description

A fast decision-making method for merging mode of 3D-HEVC interframe coding

technical field

The present invention relates to the technical field of video coding, in particular to a method for fast decision-making of an inter-frame coding combination mode in the 3D video coding standard 3D-HEVC based on the high-efficiency video coding standard HEVC.

Background technique

The High Efficiency Video Coding standard (HEVC) was jointly established by the International Organization for Standardization/International Electrotechnical Commission (ISO/IEC) Moving Picture Experts Group (MPEG) and the International Telecommunication Union Telecommunication Standardization Organization (ITU-T) Video Coding Experts Group (VCEG). A new generation of video coding standard formulated by the Joint Video Coding Working Group (JCT-VC), its official name in ISO/IEC is the second part of the MPEG-H standard, and its official name in ITU-T is H.265 . With the continuous improvement and development of HEVC, MPEG and VCEG established the 3D Video Coding Joint Working Group (JCT-3V) in 2012 to conduct research on the 3D video coding extension of HEVC, aiming to formulate HEVC-based 3D video coding standards ( 3D-HEVC). At present, 3D-HEVC has been discussed, researched and improved in 10 JCT-3V conferences, and various technologies are becoming more and more mature. 3D-HEVC includes the key technologies of HEVC, and based on the characteristics of 3D video, new coding tools are added to improve the coding efficiency of 3D video.

In 3D-HEVC, 3D video is represented in the form of multi-view texture video plus depth, and all texture maps and depth maps contained in it are encoded in sequence in the order of processing units. Each processing unit consists of several sets of texture maps and depth maps from different viewpoints at the same moment, as shown in Figure 1. In one processing unit, texture maps and depth maps for independent viewpoints are first encoded. The viewpoint adopts an independent encoding method, which is encoded by the HEVC encoder, and does not depend on the information of other viewpoints. Then, the texture and depth maps of the dependent views are encoded. When the viewpoint is coded, it refers to the information of the coded independent viewpoint to predict the information of the current viewpoint, which reduces the redundancy between viewpoints and ensures the coding efficiency. After the above process is completed, the encoding of the current processing unit is completed. For texture maps with non-independent viewpoints, each coding unit (CU) inside needs to perform inter-frame coding and intra-frame coding in sequence, and then select the coding mode with the smallest rate-distortion cost based on the rate-distortion model as the best coding mode for the current CU. best encoding mode. Among them, the rate-distortion cost is calculated by J=D+λ·R, D represents the sum of squared differences between the original image block and the reconstructed image block, λ represents the Lagrangian multiplier, and R represents the required The total number of encoded bits, including information such as encoding mode and residual coefficient. For inter-frame coding, in order to flexibly match the image content and obtain the best coding effect, it contains a total of 8 CU-based prediction unit (PU) partition modes, and the partition sizes are: 2N×2N, 2N×N, N ×2N, N×N, 2N×nU, 2N×nD, nL×2N and nR×2N, as shown in Figure 2. When performing inter-frame coding, you first need to check the merge mode, merge-skip mode, and inter-frame prediction mode in the 2N×2N partition mode, and then check the remaining 7 CU-based PU partition modes (wherein the NxN mode is only for size It is valid for 8×8 CU) to check the inter prediction mode and merge mode in turn, and at most 17 coding modes need to be checked. For intra coding, it includes two CU-based PU division modes: 2N×2N and N×N. When doing intra coding, you need to check the intra prediction mode in these 2 PU partition modes. In addition, if the CU size allowed by the PCM mode is met, the PCM mode needs to be checked. Therefore, when encoding a texture map of a non-independent viewpoint, each CU needs to check at most 20 encoding modes to obtain the optimal encoding mode, and the encoding complexity is extremely high. In the actual coding process of non-independent viewpoints, 94.0% of the CUs in the average texture map of each frame choose the merge mode as the best coding mode. Therefore, how to quickly determine the merging mode is very important to reduce the coding complexity of the dependent view texture map.

At present, 3D-HEVC adopts a fast determination method of merging mode based on the reference block of the viewpoint direction. This method uses the best coding mode of the reference block in the view direction to predict the coding mode of the current CU. When the corresponding block of the current CU in the view direction and its adjacent blocks all select the merge mode as the best coding mode, and in the 2N×2N partition mode, the rate-distortion cost obtained by using the merge-skip mode encoding of the current CU is better than using the merge mode. When the rate-distortion cost obtained by mode encoding, the current CU only checks the merge mode in all PU partition modes, and skips other encoding modes. This method can reduce the coding complexity of B frames to a certain extent, but because it does not make full use of the redundant information between viewpoints, it can only cover a small part of CUs in each frame of images, resulting in most of them choosing the merge mode as the best coding mode The CU still needs to check all the coding modes to get the best coding mode, and the coding efficiency still needs to be improved. For a P frame, since the reference frame in the view direction is an I frame, which is coded in intra-frame prediction mode, all CUs in the texture map that use the merge mode as the best coding mode do not meet the fast judgment conditions of the merge mode, and this method fails , the coding complexity of P frames is not improved.

Based on 3D-HEVC, the present invention proposes a fast decision-making method for inter-coding merging modes based on inter-view coding mode correlation and CU partition layer coding mode correlation, and improves the choice of merging mode as the best under the premise of ensuring coding quality. The prediction rate of the CU in the encoding mode can effectively reduce the encoding complexity.

Contents of the invention

The technical problem to be solved by the present invention is to solve the problem of high coding complexity of the image frame of the 3D video non-independent viewpoint texture map, and to provide a fast judgment by the correlation between the coding modes between the viewpoints and the coding mode between the CU division levels The method of merging modes can effectively reduce the coding complexity while ensuring the coding quality.

The technical solution adopted by the present invention to solve the above technical problems is: a fast decision-making method for merging modes of 3D-HEVC inter-frame coding, and a fast algorithm is designed for 3D video non-independent viewpoint texture maps. The flow chart of the technical solution of the present invention is shown in FIG. 3 . Define J _skip (CU _cur ) to indicate the rate-distortion cost obtained by encoding the current CU in the merge-skip mode, and J _merge (CU _cur ) to indicate the rate-distortion cost obtained by encoding the current CU in the merge mode. Method of the present invention comprises the steps:

1) Extract the CU to be coded from the texture map of the non-independent view.

2) The current CU checks merge mode and merge-skip mode in 2N×2N partition mode.

3) Determine whether the frame where the current CU is located is a P frame or a B frame.

4) Through the judgment of the previous step, if the obtained result is a P frame, judge whether the division level of the current CU is 3;

If the division level of the current CU is 3, that is, the size of the current CU is 8×8, determine whether condition a and condition b are satisfied at the same time:

Condition a. The merge mode is used as the best coding mode for the CU at the previous split level.

Condition b. In the 2N×2N division mode, J _skip (CU _cur )<J _merge (CU _cur ).

If condition a and condition b hold at the same time, the current CU only checks the merge mode in 2N×N, N×2N, N×N, 2N×nU, 2N×nD, nL×2N and nR×2N partition modes, skip other encoding modes; otherwise, the current CU checks all encoding modes.

If the division level of the current CU is 0-2, that is, when the size of the current CU is 64×64, 32×32 or 16×16, all coding modes are checked.

5) Through the judgment of step 3, if the obtained result is a B frame, determine the five reference blocks of the current CU in the direction of the viewpoint through the disparity vector (DoNBDV) obtained from the adjacent blocks, including the corresponding block of the current CU in the direction of the viewpoint and its 4 adjacent blocks. Determine whether the current CU satisfies condition c and condition d at the same time:

Condition c. Any reference block in the reference frame in the view direction adopts the merge mode as the best coding mode.

Condition d. In the 2N×2N division mode, J _skip (CU _cur )<J _merge (CU _cur ).

If condition c and condition d are true at the same time, the current CU only checks the merge mode in 2N×N, N×2N, N×N, 2N×nU, 2N×nD, nL×2N and nR×2N partition modes, skip other encoding modes; otherwise, the current CU checks all encoding modes.

Compared with the prior art, the present invention has the advantage that: the traditional method for quickly determining the merge mode determines the encoding mode of the current CU based on the best encoding mode of the reference block in the view direction, only when five reference blocks adopt the merge mode at the same time as the best When the encoding mode is selected, it can be determined that the best encoding mode of the current CU is merge mode, and the redundant information between views is not fully utilized; for P frames, the CU whose best encoding mode is merge mode cannot be determined through the view direction reference block, the method invalidated. The present invention makes full use of the inter-viewpoint correlation and the inter-division level correlation of the CU coding mode to predict the CU that adopts the merge mode as the best coding mode. For a B frame, use the best coding mode of the corresponding block of the current CU in the reference frame in the view direction and its adjacent blocks to determine the CU whose best coding mode is merge mode; for a P frame, use the best coding mode of the CU located in the previous division The best coding mode determines the CU whose best coding mode under the current division level is merge mode. This method can reduce the coding complexity of the non-independent viewpoint texture map and improve the coding speed while maintaining the coding quality.

Experimental results show that, compared with the fast determination method of combining modes used in 3D-HEVC, the method of the present invention can save an average of 20.4% of non-independent viewpoint texture maps with only a small increase in code stream and basically no decrease in coding quality encoding time.

Description of drawings

FIG. 1 is a schematic diagram of a coding structure of a 3D video processing unit;

FIG. 2 is a schematic diagram of a PU partition mode based on a CU;

Fig. 3 is a flow chart of the inventive method;

FIG. 4 is a schematic diagram of a hierarchical structure of CU division;

FIG. 5 is a schematic diagram of a reference block in a view direction reference frame of the current CU.

detailed description

The present invention will be described in further detail below in conjunction with the accompanying drawings.

The present invention designs a fast decision-making method for an inter-frame coding merging mode for a CU of a 3D video non-independent viewpoint texture map. In actual use, the encoder will call the method of the present invention to complete specific encoding work. Fig. 3 is a flowchart of the method of the present invention. Define J _skip (CU _cur ) to indicate the rate-distortion cost obtained by encoding the current CU in the merge-skip mode, and J _merge (CU _cur ) to indicate the rate-distortion cost obtained by encoding the current CU in the merge mode. Method steps of the present invention are as follows:

Step 1: Read in the video sequence according to the configuration file of the encoder, and configure the encoder through the parameter information in the configuration file.

Step 2: sequentially extract the CUs to be coded from the texture map of the non-independent viewpoint.

Step 3: The current CU checks merge mode and merge-skip mode in 2N×2N partition mode.

Step 4: Determine whether the frame where the current CU is located is a P frame or a B frame.

Step 5: Judging by the previous step, if the obtained result is a P frame, judge whether the division level of the current CU is 3;

1) If the division level of the current CU is 3, that is, when the size of the current CU is 8×8, determine whether to

Condition a and condition b are met:

Condition a. The merge mode is used as the best coding mode for the CU at the previous split level.

Condition b. J _skip (CU _cur )<J _merge (CU _cur ) in the 2N×2N division mode.

Among them, the CU of the previous division level mentioned in condition a means that the CU of the size of the current CU is 16×16 among the four 8×8 CUs obtained after the division of the current level, which is located at the previous division level, As shown in Figure 4. If condition a and condition b hold at the same time, the current CU only checks the merge mode in 2N×N, N×2N, N×N, 2N×nU, 2N×nD, nL×2N and nR×2N partition modes, skip For other coding modes, the coding mode with the smallest rate-distortion cost is selected as the best coding mode; otherwise, the current CU checks all coding modes, and the coding mode with the smallest rate-distortion cost is selected as the best coding mode. 2) If the division level of the current CU is 0~2, that is, the size of the current CU is 64×64, 32×32 or 16×16

When , all coding modes are checked, and the coding mode with the smallest rate-distortion cost is selected as the best coding mode.

Step 6: Through the judgment of the fourth step, if the obtained result is a B frame, determine the five reference blocks of the current CU in the direction of the viewpoint through the disparity vector (DoNBDV) obtained from the adjacent block, including the current CU in the direction of the viewpoint The corresponding block and its adjacent left block, upper block, right block and lower block are shown in Figure 5. Wherein, the corresponding block of the current CU in the view direction is the disparity compensation block of the current CU in the view direction reference frame obtained by DoNBDV. Determine whether the current CU satisfies condition c and condition d at the same time:

Condition c. Any reference block in the reference frame in the view direction adopts the merge mode as the best coding mode.

Condition d. J _skip (CU _cur )<J _merge (CU _cur ) in the 2N×2N division mode.

If condition c and condition d are true at the same time, the current CU only checks the merge mode in 2N×N, N×2N, N×N, 2N×nU, 2N×nD, nL×2N and nR×2N partition modes, skip For other coding modes, the coding mode with the smallest rate-distortion cost is selected as the best coding mode; otherwise, the current CU checks all coding modes, and the coding mode with the smallest rate-distortion cost is selected as the best coding mode.

In order to test the performance of the method proposed in the present invention, the method of the present invention is compared with the fast judgment method of the merge mode in 3D-HEVC. The experimental platform is HTM-10.0, and the test sequences are Balloons, Kendo, Newspaper, GT_Fly, Pozan_Hall2 and Poznan_Street sequences, including two resolutions of 1024×768 and 1920×1088. Four QP combinations (25, 34), (30, 39), (35, 42) and (40, 45) are used. A total of 3 viewpoints are encoded, including 1 independent viewpoint (reference viewpoint) and 2 dependent viewpoints (test viewpoint), and each viewpoint encodes a video with a length of 10s. The experimental conditions and experimental parameters are carried out in accordance with the general test conditions published by JCT-3V. All experiments were performed independently on a PC configured with an Intel(R) Core(TM) i7-3770 CPU and 8GB RAM.

Table 1 shows the coding performance comparison results between the method of the present invention and the method adopted in 3D-HEVC. It can be seen from the table that the method of the present invention has only an average BD-Rate increase of 0.3%, 0.3% and 0.0% compared with the fast determination method of the merge mode in 3D-HEVC, and basically no decline in encoding quality can be seen. And the method of the present invention can save 6.5% of the total coding time and 20.4% of the coding time of the dependent viewpoint texture map on average.

Table 1 Comparison of encoding performance results of different methods

Claims (4)

1. A fast decision-making method for merging mode of 3D-HEVC inter-frame coding, defining J _skip (CU _cur ) to represent the rate-distortion cost obtained by merging-skip mode encoding of the current CU, and J _merge (CU _cur ) to represent the merging of the current CU The rate-distortion cost obtained by mode encoding is characterized in that it includes the following steps, 1) Extract the CUs to be coded in order from the texture map of the non-independent viewpoint; 2) The current CU checks the merge mode and the merge-skip mode in the 2N×2N partition mode; 3) Determine whether the frame where the CU is located is a P frame or a B frame; 4) Through the judgment of the previous step, if the obtained result is a P frame, judge whether the division level of the current CU is 3; If the division level of the current CU is 3, that is, the size of the current CU is 8×8, determine whether condition a and condition b are satisfied at the same time: Condition a. The CU of the previous division level adopts the merge mode as the best coding mode; Condition b. In the 2N×2N division mode, J _skip (CU _cur )<J _merge (CU _cur ); If condition a and condition b hold at the same time, the current CU only checks the merge mode in 2N×N, N×2N, N×N, 2N×nU, 2N×nD, nL×2N and nR×2N partition modes, skip For other coding modes, select the coding mode with the smallest rate-distortion cost as the best coding mode; otherwise, the current CU checks all coding modes, and select the coding mode with the smallest rate-distortion cost as the best coding mode; If the division level of the current CU is 0 to 2, that is, when the size of the current CU is 64×64, 32×32 or 16×16, check all coding modes, and select the coding mode with the smallest rate-distortion cost as the best coding mode; 5) Through the judgment in step 3, if the obtained result is a B frame, determine the five reference blocks of the current CU in the viewpoint direction through the disparity vector obtained from the adjacent block; judge whether the current CU satisfies condition c and condition d at the same time: Condition c. Any reference block in the reference frame in the view direction adopts the merge mode as the best coding mode; Condition d. In the 2N×2N division mode, J _skip (CU _cur )<J _merge (CU _cur ); If condition c and condition d are true at the same time, the current CU only checks the merge mode in 2N×N, N×2N, N×N, 2N×nU, 2N×nD, nL×2N and nR×2N partition modes, skip For other coding modes, the coding mode with the smallest rate-distortion cost is selected as the best coding mode; otherwise, the current CU checks all coding modes, and the coding mode with the smallest rate-distortion cost is selected as the best coding mode. 2. A fast decision-making method for 3D-HEVC inter-frame coding merging mode according to claim 1, characterized in that: in step 4, the CU at the previous division level indicates that it is located at the previous division level and satisfies the requirements of the division at the current level. The four obtained CUs with a size of 8×8 include a CU with a size of 16×16 of the current CU. 3. A fast decision-making method for 3D-HEVC inter-frame coding merging mode according to claim 1, characterized in that: in the step 5, the five reference blocks of the current CU in the direction of the viewpoint include the corresponding blocks of the current CU in the direction of the viewpoint block and its adjacent left block, upper block, right block and lower block; wherein, the corresponding block of the current CU in the view direction is the disparity compensation block of the current CU in the view direction reference frame obtained by DoNBDV. 4. a kind of 3D-HEVC inter-frame encoding merge mode fast decision-making method according to claim 1 is characterized in that: in actual use, encoder will call the method in the present invention to complete specific encoding work; definition J _skip (CU _cur ) indicates the rate-distortion cost obtained by encoding the current CU using the merge-skip mode, and J _merge (CU _cur ) indicates the rate-distortion cost obtained by encoding the current CU using the merge mode; the steps are as follows, Step 1: Read in the video sequence according to the configuration file of the encoder, and configure the encoder through the parameter information in the configuration file; Step 2: Extract the CUs to be coded sequentially from the texture map of the non-independent viewpoint; Step 3: The current CU checks the merge mode and the merge-skip mode in the 2N×2N partition mode; Step 4: Determine whether the frame where the current CU is located is a P frame or a B frame; Step 5: Judging by the previous step, if the obtained result is a P frame, judge whether the division level of the current CU is 3; 1) If the division level of the current CU is 3, that is, when the size of the current CU is 8×8, determine whether condition a and condition b are satisfied at the same time: Condition a. The CU of the previous division level adopts the merge mode as the best coding mode; Condition b. J _skip (CU _cur )<J _merge (CU _cur ) in 2N×2N division mode; If condition a and condition b hold at the same time, the current CU only checks the merge mode in 2N×N, N×2N, N×N, 2N×nU, 2N×nD, nL×2N and nR×2N partition modes, skip For other coding modes, select the coding mode with the smallest rate-distortion cost as the best coding mode; otherwise, the current CU checks all coding modes, and select the coding mode with the smallest rate-distortion cost as the best coding mode; 2) If the division level of the current CU is 0~2, that is, when the size of the current CU is 64×64, 32×32 or 16×16, check all coding modes, and select the coding mode with the smallest rate-distortion cost as the best coding mode ; Step 6: Through the judgment of the fourth step, if the obtained result is a B frame, determine the 5 reference blocks of the current CU in the viewpoint direction through the disparity vector obtained from the adjacent block; judge whether the current CU satisfies the condition c and the condition at the same time d: Condition c. Any reference block in the reference frame in the view direction adopts the merge mode as the best coding mode; Condition d. J _skip (CU _cur )<J _merge (CU _cur ) in 2N×2N division mode; If condition c and condition d are true at the same time, the current CU only checks the merge mode in 2N×N, N×2N, N×N, 2N×nU, 2N×nD, nL×2N and nR×2N partition modes, skip For other coding modes, the coding mode with the smallest rate-distortion cost is selected as the best coding mode; otherwise, the current CU checks all coding modes, and the coding mode with the smallest rate-distortion cost is selected as the best coding mode.