Fast VCC coding unit division method based on spatiotemporal depth information

Abstract

The present invention proposes a VCC coding unit fast division method based on spatiotemporal depth information, the steps are: quadtree division of the CU of the maximum coding unit, multi-type tree division after the quadtree division; when the quadtree division depth of the maximum coding unit is 2, the average depth of spatiotemporal adjacent CUs is less than 2, the quadtree division is terminated; when the quadtree division depth of the maximum coding unit is 3, the average depth of spatiotemporal adjacent CUs is less than 3, the quadtree division is terminated; when the quadtree division depth of the maximum coding unit is between 2 and 4, the CU is divided into multi-type trees; when the multi-type tree division depth of the maximum coding unit is 0, the average depth of spatiotemporal adjacent CUs is less than 0.1, the CU does not need multi-type tree division, otherwise all division modes are traversed; the optimal prediction method is determined according to the rate distortion cost. The present invention can save an average of 35.72% of the encoding time, and the average bit rate increases less under the same encoding quality.

Description

Technical Field

[0001] The present invention relates to the technical field of video coding, and in particular to a method for quickly dividing a VCC coding unit based on spatiotemporal depth information. Background Art

[0002] As people's demand for high-definition video increases, more and more video coding standards have emerged, such as H.264 / AVC and H.265 / HEVC. However, the compression ratios that these standards can achieve cannot keep up with the rapid growth of video data demand. Therefore, the research and standard formulation of the next generation of ultra-high-definition video coding technology has become very urgent. The video coding standard H.265 / HEVC, which is widely used in the video field, gradually cannot meet the needs of practical applications. In this case, the new generation of video coding standard H.266 / VVC came into being.

[0003] HEVC is the predecessor of versatile video coding (VVC). HEVC uses a quadtree coding structure. In the past decade, a lot of research has helped reduce the complexity of HEVC. In HEVC, the coding unit (CU) partition takes up the largest proportion of the coding time, so many methods have tried to simplify the CU partition to reduce the complexity of HEVC. Similarly, the CU partition structure of VVC can also be simplified. The CU partition structure of VVC is much more flexible than that of HEVC. It uses a more complex quadtree, tritree, and binary tree coding structure, and the amount of calculation is much larger. A large number of fast partitioning algorithms have been proposed, which can be roughly divided into two categories: heuristic methods and machine learning methods.

[0004] In the heuristic method, by utilizing some intermediate features of coding, such as texture homogeneity / complexity and spatial correlation, a statistical model of CU partition is established. Through these models, further CU partitioning is terminated in advance, ending the redundant rate-distortion optimization (RDO) process in CU partitioning. Yang et al. proposed a low-complexity code tree unit (CTU) structure decision algorithm based on global and local texture information, context information, etc. Wang et al. also described a local constrained CU partition decision method. In addition to texture analysis, some works are also based on machine learning or convolutional neural networks. Amestoy et al. introduced random forests into VVC frame partitioning. The application of CNN (convolutional neural network) in video coding has been effectively verified in previous works. Liu et al. proposed a CNN-based CU mode and partition decision for HEVC intra-frame coding. Xu et al. proposed a method of applying CNN and long short-term memory (LSTM) network to HEVC intra- and intra-partitioning to reduce coding complexity. Also in VVC, some works similar to CNN have appeared. Jin et al. proposed a fast quadtree plus binary tree (QTBT) decision for CNN for VVC intra prediction. This work models the QTBT structure as a classification problem. For 32x32 CUs, the CU depth will be inferred by CNN, where the CU depth is defined by the quadtree depth and the binary tree depth, which is different from the depth in HEVC. Tissier et al. proposed a CNN-based intra-coding complexity reduction technology, which obtains a probability vector through CNN and obtains the CU partitioning method based on these probability vectors, thereby accelerating the encoding speed. Chen et al. proposed an algorithm that uses the average depth of the coding unit to accelerate the early termination of CU partitioning. The above methods are all carefully designed for fast coding unit partitioning, which greatly shortens the encoding time, but significantly reduces the performance of RD.

[0005] Therefore, in the new generation of video coding standards, in versatile video coding (VVC), the characteristics of coding units (CU) of different sizes have significantly improved the coding efficiency, but at the same time brought extremely high computational complexity. Summary of the invention

[0006] In view of the technical problem of high computational complexity of existing VCC methods, the present invention proposes a VCC coding unit fast partitioning method based on spatiotemporal depth information, which removes redundant calculations in CU partitions. Without affecting the coding performance, it has low computational complexity, reduces unnecessary rate-distortion optimization operations, and significantly speeds up the coding speed.

[0007] In order to achieve the above object, the technical solution of the present invention is implemented as follows: a VCC coding unit fast division method based on spatiotemporal depth information, the steps of which are as follows:

[0008] S1: The CU of the largest coding unit in VVC is divided into quadtrees. After the quadtree division is completed, multi-type tree division is performed to perform correlation analysis between temporally and spatially adjacent blocks;

[0009] S2: When the quadtree partition depth of the maximum coding unit is 2, if the average depth of the spatiotemporally adjacent CUs is less than 2, the quadtree partition test process is terminated; otherwise, it proceeds to step S6;

[0010] S3: When the quadtree partition depth of the maximum coding unit is 3, if the average depth of the spatiotemporally adjacent CUs is less than 3, the quadtree partition test process is terminated; otherwise, it proceeds to step S6;

[0011] S4: When the quadtree partition depth of the maximum coding unit is between 2 and 4, the CU performs a multi-type tree partition test process;

[0012] S5: When the multi-type tree division depth of the largest coding unit is 0, if the average depth of the spatiotemporally adjacent CUs is less than 0.1, the CU does not need to perform the multi-type tree division test process, otherwise all division modes are traversed;

[0013] S6: Determine the optimal prediction method according to the rate-distortion cost: calculate the rate-distortion cost for all the partition modes of the current CU, and select the partition mode with the smallest rate-distortion cost as the partition mode of the current CU.

[0014] The spatiotemporally adjacent CUs include the left coding unit L-CU, the upper coding unit U-CU and the coding unit Col-CU at the same position as the current CU in the previous frame of the current CU; Col-CU, L-CU and U-CU are used to predict the depth of the current CU, and equal weights are assigned to each spatiotemporally adjacent CU.

[0015] The quadtree partition depth of the maximum coding unit is represented by the depth mean of the maximum coding block: a maximum coding unit is divided into 4 CUs, the depth mean of the 4 CUs is calculated respectively, the depth values ​​of the 4 CUs are added together to obtain the average value, and the depth of the maximum coding unit is obtained.

[0016] The relationship between the average depth of the quadtree partition and the current maximum coding unit is as follows:

[0017] Step 1. For a maximum coding unit with optimal RDO coding, calculate the number of blocks with different depths and obtain the CU with the best depth;

[0018] Step 2. Get the depth mean of three spatiotemporal adjacent blocks Col-CU, L-CU and U-CU respectively;

[0019] Step 3. Take the weighted sum of the quadtree partition depth means of the coding block Col-CU of the previous frame, the coded left coding block L-CU, and the upper coding block U-CU.

[0020] The method for determining whether to further divide the current CU for RDO testing based on the calculated depth of the quadtree division is:

[0021] (1) When the test quadtree partition depth is 0 or 1, the CU is partitioned for further RDO testing;

[0022] (2) When the test quadtree partition depth is 2, calculate the average quadtree partition depth of the spatiotemporal adjacent blocks Col-CU, L-CU and U-CU. If the depth averages are all less than 2 and the quadtree partition test depth of the current CU is 2, the quadtree partition is terminated in advance;

[0023] (3) When the test quadtree partition depth is 3, the average quadtree partition depth of the spatiotemporally adjacent blocks Col-CU, L-CU and U-CU is calculated. If the depth averages are all less than 3, and the quadtree partition test depth of the current coding unit is 3, the quadtree partition is terminated in advance.

[0024] The analysis method of the correlation between the multi-type tree partition depth information of the current CU and the spatiotemporal neighboring blocks Col-CU, L-CU and U-CU is:

[0025] 1). When the optimal quadtree partition depth of the maximum coding unit is 2, the optimal multi-type tree partition depth is 0, 1, 2, and 3 blocks respectively; when the optimal quadtree partition depth of the maximum coding unit is 3, the optimal multi-type tree (MTT) partition depth is 0, 1, 2, and 3 blocks respectively; when the optimal quadtree partition depth of the maximum coding unit is 4, the optimal multi-type tree partition depth is 0, 1, and 2 blocks respectively;

[0026] 2) Calculate the mean of the multi-type tree partition depth of the spatiotemporal neighboring blocks Col-CU, L-CU and U-CU of each maximum coding unit;

[0027] 3) The multi-type tree partition depth mean values ​​of the obtained spatiotemporal adjacent blocks Col-CU, L-CU and U-CU are assigned equal weights for calculation.

[0028] The beneficial effects of the present invention are as follows: First, the spatiotemporal correlation existing in the video is utilized to predict the coding depth of the current CU according to the depth mean of the spatiotemporally adjacent CUs; ​​then, unnecessary traversal ranges are skipped during encoding to achieve fast division of coding units to save encoding time. The present invention makes full use of the spatiotemporal correlation of video sequences, and through average depth analysis at the maximum coding block level, and by comparing the average depth and threshold of spatiotemporally adjacent blocks, it can reduce unnecessary coding block division and RDO (redundancy rate distortion optimization) testing. Experimental results show that compared with the original encoder, the present invention can save an average of 35.72% of encoding time, and the average bit rate (BDBR) increases by only 0.43% at the same encoding quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 It is a schematic diagram of the process of the present invention.

[0031] Figure 2 Schematic diagram of spatiotemporally adjacent CUs of the present invention, where (a) is the previous frame and (b) is the current frame.

[0032] Figure 3 Schematic diagram of calculating the average coding depth of four adjacent CUs according to the present invention.

[0033] Figure 4 Schematic diagram of calculating the average coding depth of the largest coding block according to the present invention.

[0034] Figure 5 It is a schematic diagram of maximum coding block depth analysis of the present invention.

[0035] Figure 6 A schematic diagram of calculating the mean depth of multiple types of partitions of adjacent blocks according to the present invention.

[0036] Figure 7 It is a schematic diagram of calculating the mean depth of quadtree partitioning of adjacent blocks according to the present invention.

[0037] Figure 8 It is a schematic diagram of calculating the depth mean of adjacent blocks according to the present invention.

[0038] Fig. 9 A schematic diagram of depth calculation of multi-type tree partitioning of a maximum coding block according to the present invention.

[0039] Fig.10This is a comparison diagram of the BDBR increase between the present invention and the G. Tang and Tissier methods.

[0040] Fig.11 This is a comparison result diagram of TS increase between the present invention and the G. Tang and Tissier methods.

[0041] Fig.12 Schematic diagram of RD curves of the present invention and the VTM7.0 algorithm at different QPs. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] like Figure 1 As shown, the embodiment of the present invention proposes a VVC coding unit fast division method based on spatiotemporal depth information, which aims at the problem that the quadtree nested multi-type tree (MTT) division method of CU used in VVC causes a significant increase in the complexity of the encoder. The spatiotemporal correlation existing in the video is used to predict the coding depth of the current CU according to the depth mean of the spatiotemporal adjacent CUs. The present invention proposes a fast QT decision algorithm and a fast MTT decision algorithm, and the main process of the algorithm is as follows:

[0044] S1: The largest coding unit (LCU) in VVC has a size of 128×128 and a depth of 0. The depth of CU ranges from 0 to 6. CU is first divided into quadtree (QT). After the quadtree (QT) division is completed, multi-type tree (MTT) division is performed. We first perform correlation analysis between adjacent blocks in time and space.

[0045] S2: Because the depth level between adjacent CUs may not change significantly, when the quadtree partitioning depth of the maximum coding unit is 2, if the average depth of the three temporally and spatially adjacent blocks (Col-CU, L-CU and U-CU) is less than 2, the quadtree partitioning test process is terminated early.

[0046] S3: Because the depth level between adjacent CUs may not change significantly, when the quadtree partition depth of the maximum coding unit is 3, if the average depth of the three temporally and spatially adjacent blocks (Col-CU, L-CU and U-CU) is less than 3, the quadtree partition test process is terminated early.

[0047] S4: When the quadtree partition depth of the largest coding unit is between 2 and 4 (the QT depth of the current CU is different, and the average depth of adjacent blocks to be judged is also different), the coding unit needs to perform an additional multi-type tree partition test process.

[0048] S5: When the multi-type tree division depth of the largest coding unit is 0, if the average depth of the three temporally and spatially adjacent blocks (Col-CU, L-CU and U-CU) is less than 0.1 (the division depth cannot be less than 0. When the average depth of adjacent CUs is less than 0.1, it means that the depth of the current CU is close to the depth of the adjacent CU), the coding unit does not need to perform the next multi-type tree division test process, otherwise all division modes are traversed.

[0049] S6: Determine the optimal prediction method according to the rate-distortion cost RDCost, try all the partition modes of the current CU and calculate their RDCost, and select the partition mode with the minimum cost as the partition mode of the current CU.

[0050] In natural images, adjacent CUs usually have similar textures, and the optimal depth level of the current CU is strongly correlated with its adjacent CUs, and the depth level between adjacent CUs may not change significantly. Figure 2 As shown in the figure, Cur-CU represents the current CU, L-CU, LU-CU, U-CU and RU-CU represent the left CU, upper left CU, upper CU and upper right CU of the current CU respectively, and Col-CU represents the CU at the same position as the current CU in the previous frame. In order to analyze the correlation between the depth information of adjacent CUs in time and space. Taking LCU as the unit, 12 videos are selected from the standard test video sequences ClassB~ClassE of VVC, with quantization parameters (QP) of 22, 27, 32, and 37, and full I frame encoding is performed. When the depths of L-CU, LU-CU, U-CU, RU-CU and Col-CU are d, the probability that Cur-CU is at the same depth d at the same time is used to describe their correlation.

[0051] According to the correlation of the depths of spatiotemporally adjacent LCUs, LU-CU and RU-CU are excluded, and finally Col-CU, L-CU and U-CU are selected to predict the depth of Cur-CU, assigning equal weights to each adjacent LCU.

[0052] First, the QT coding acceleration method is used. Considering that the depth information of the adjacent coding blocks divided by the quadtree (QT) is very complex, in order to simplify the calculation process, the depth mean of the largest coding block is used to represent it. The specific calculation steps are as follows Figure 3 and Figure 4 As shown. Figure 3As shown, first, an LCU (largest coding unit) is divided into 4 small coding blocks, and then the depth mean of the 4 CUs is calculated respectively to obtain four sub-blocks with depth values ​​of 2, 2.25, 2.25 and 1. Finally, the depth values ​​of the 4 CUs are added and averaged to obtain the depth of the entire LCU as 1.875, as shown in Figure 4 shown.

[0053] Statistical analysis of the relationship between the average depth of quadtree partitioning and the current LCU. The detailed statistical steps are as follows:

[0054] Step 1. For an LCU with optimal RDO coding, calculate the number of blocks with different depths, and finally get the CU with the best depth. Figure 5 In the example shown, “C” represents the depth of the current CU, which includes three depth 1 blocks, three depth 2 blocks, and one depth 3 block.

[0055] Step 2. Through the previous analysis, it is found that there is a strong correlation between the current largest coding unit and the coding block of the previous frame and the surrounding coded left coding block and upper coding block. The depth mean of these three spatiotemporal adjacent blocks is obtained respectively, such as Figure 5 shown.

[0056] Step 3: Take the weighted sum of the depth mean of the quadtree (QT) partition of the coding block of the previous frame, the surrounding coded left coding block and the upper coding block.

[0057] Based on the depth information calculated by the above statistical data, the encoder can decide whether to further divide the current coding unit for RDO testing. The detailed operation is as follows:

[0058] (1) When the test quadtree (QT) partition depth is 0 or 1, due to the large size of the coding unit, the probability of the coding unit with the best depth of 0 and 1 is very small compared with other coding unit depths. When testing these two depths, we usually split the coding unit for further RDO testing.

[0059] (2) When the test quadtree (QT) partition depth is 2, the average of the quadtree (QT) partition depths of the coding block of the previous frame, the surrounding coded left coding block, and the upper coding block is calculated to determine whether to skip the RDO calculation of the coding unit. If their average depth is less than 2, the probability that the LCU is finally divided into a depth of 0 to 2 is relatively high.

[0060] (3) When the test quadtree (QT) partition depth is 3, the average of the quadtree (QT) partition depths of the coding block of the previous frame, the surrounding coded left coding block, and the upper coding block is calculated to determine whether to skip the RDO calculation of the coding unit. If their average depth is less than 3, it is very likely that the final LCU depth is 0 to 3.

[0061] In order to prove the above two relationships (2) and (3), experiments were conducted under the same settings and test videos. For relationship (2), when the average depth of adjacent blocks (L-CU, U-CU and Col-CU) is less than or equal to 2, the probability that the optimal depth of the current coding unit is less than or equal to 2. The same method is used to prove relationship (3). The results show that the accuracy of relationship (2) and relationship (3) in percentage is higher than 80%, which also confirms the feasibility of these two relationships. These two relationships can help reduce unnecessary block division and RDO processes and speed up the encoding process. The specific steps are as follows:

[0062] Step 1. If the average depths of the coding block at the same position of the previous frame, the surrounding coded left coding block and the upper coding block are all less than 2, and the quadtree (QT) partition test depth of the current coding unit is 2, the quadtree (QT) partition is terminated in advance.

[0063] Step 2: If the average depths of the coding block at the same position of the previous frame, the surrounding coded left coding block and the upper coding block are all less than 3, and the quadtree (QT) partition test depth of the current coding unit is 3, the quadtree (QT) partition is terminated in advance.

[0064] The encoding acceleration process of quadtree (QT) partitioning was introduced in detail above. The encoding acceleration method of multi-type tree (MTT) partitioning is similar to that of quadtree (QT) partitioning. Figure 6-8 As shown, first, a maximum coding block is divided into 4 CUs, and the 4 individual CUs are further divided into 4 sub-blocks, and the multi-type tree (MTT) partition depth values ​​of the 16 sub-blocks are calculated, as shown in FIG. Figure 6 As shown, the multi-type tree (MTT) of these 16 sub-blocks is then used to divide the depth values, and the average depth values ​​of the four individual CUs are calculated to obtain four sub-blocks with depth values ​​of 0.625, 0.5, 0.625, and 0, as shown in FIG. Figure 7 Finally, the average depths of the four CUs are added together to obtain the average depth of the largest coding block, which is 0.4375, as shown in Figure 8 shown.

[0065] How to analyze the correlation between the current coding unit and the coding block of the previous frame, the surrounding coded left coding block, and the multi-type tree (MTT) partition depth information of the upper coding block. The detailed operations are as follows:

[0066] Step 1. Calculate the number of blocks with optimal multi-type tree (MTT) partition depths of 0, 1, 2, and 3 when the optimal quadtree (QT) partition depth of LCU is 2. Calculate the number of blocks with optimal multi-type tree (MTT) partition depths of 0, 1, 2, and 3 when the optimal quadtree (QT) partition depth of LCU is 3. Calculate the number of blocks with optimal multi-type tree (MTT) partition depths of 0, 1, and 2 when the optimal quadtree (QT) partition depth of LCU is 4. Fig. 9 As shown, “C” represents the current CU, which consists of multi-type trees (MTTs) with depths of 0, 1, and 2.

[0067] Step 2. Calculate the average MTT depth of each LCU adjacent to the LCU block. Fig. 9 As shown, the multi-type tree (MTT) partition depth means of the coding block Col-CU of the previous frame, the surrounding coded left coding block L-CU and the upper coding block U-CU are calculated respectively.

[0068] Step 3. Assign equal weights to the obtained MTT depth mean of the Col-CU of the previous frame and the multi-type tree (MTT) partition depth mean of the surrounding encoded L-CU and U-CU.

[0069] After the above calculations and analysis, we found that there is a depth-based rule that can use the depth of adjacent blocks to decide whether to continue block division and RDO testing. The rule is described as follows:

[0070] When the depth of the test CU is 0, if the average value of the adjacent blocks is less than 0.1, then the final optimal MTT depth is likely to be 0. In order to confirm the proposed depth-based rule to decide whether to continue the division, the same test video and system settings are used. When the average MTT depth of the adjacent blocks is less than 0.1, the probability that the MTT depth of the test CU is 0 is tested. The results show that the accuracy of the depth-based prediction rule is not less than 80%, proving that it is feasible in reducing the temporal complexity of CU encoding in VVC. When testing the MTT depth of 0, if the average depth of the adjacent blocks is less than 0.1, the block division and RDO test can be terminated early.

[0071] In order to verify the performance of the algorithm, the above algorithm was integrated into the VVC reference implementation software VTM7.0 for reference experiments. The encoding configuration mode adopted was the full I frame (AI) mode, and the encoding performance of video QP selections of 22, 27, 32 and 37 was tested respectively. In order to analyze the performance of the algorithm, the percentage of encoding time saved ΔT and the bit rate increment ΔBDBR were used as the measurement criteria. The calculation formulas are as follows (1) to (2).

[0072]

[0073]

[0074] Among them, T proposed and BDBR proposed T represent the encoding time and bit rate of the fast algorithm of the present invention respectively; VTM7.0 and BDBR VTM7.0 Respectively represent the encoding time and average bit rate of VTM7.0.

[0075] The comparison results between the present invention and the original VTM7.0 are shown in Table 1. Compared with the standard algorithm, the present invention can save an average of 35.72% of encoding time, while the impact on the video encoding quality is very small or even negligible, and the video bit rate increases by only 0.43%. And with the increase of video resolution, the absolute values ​​of BDBR and BD-PSNR tend to decrease. The algorithm proposed in the present invention is particularly outstanding in the coding rate distortion performance, effectively improving the real-time performance of VVC encoding, and has a better effect on high-resolution video.

[0076] Table 1 Comparison of test results of the present invention and VTM7.0

[0077]

[0078] In order to evaluate the encoding performance of the present invention, Figure 8 and Fig. 9 The comparison results of the overall encoding time savings and BDBR increase of the proposed overall method and the latest H.266 / VVC fast algorithm are given respectively. Fig.10 This is an example of the RD curve of the present invention and the VTM7.0 algorithm at different QPs. The RD curves of the present invention and the VTM7.0 algorithm are basically overlapped. Therefore, the encoding performance of the present invention is not significantly lower than that of the VVC standard algorithm.

[0079] The present invention utilizes the spatiotemporal correlation of video sequences and proposes a new fast method for VVC intra-frame coding unit division based on spatiotemporal depth information, including average depth analysis at the LCU level, using the average depth and threshold comparison of spatiotemporal adjacent blocks, and reducing unnecessary coding block division and RDO testing, which is particularly prominent in coding rate distortion performance.

[0080] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A VVC coding unit fast division method based on spatiotemporal depth information, characterized in that: The steps are as follows: S1: divide the CU of the largest coding unit in VVC into quadtrees and perform correlation analysis between adjacent blocks in time and space; S2: When the quadtree partition depth of the maximum coding unit is 2, if the average depth of the spatiotemporally adjacent CUs is less than 2, the quadtree partition test process is terminated and the process proceeds to step S4; if the average depth of the spatiotemporally adjacent CUs is greater than or equal to 2, the process proceeds to step S6; S3: When the quadtree partition depth of the maximum coding unit is 3, if the average depth of the spatiotemporally adjacent CUs is less than 3, the quadtree partition test process is terminated and the process proceeds to step S4; if the average depth of the spatiotemporally adjacent CUs is greater than or equal to 3, the process proceeds to step S6; S4: When the quadtree partition depth of the maximum coding unit is between 2 and 4, the CU performs a multi-type tree partition test process; S5: When the multi-type tree division depth of the maximum coding unit is 0, if the average depth of the spatiotemporally adjacent CUs is less than 0.1, the CU does not need to perform the multi-type tree division test process; if the average depth of the spatiotemporally adjacent CUs is less than 0.1, proceed to step S6; S6: Determine the optimal prediction method according to the rate-distortion cost: calculate the rate-distortion cost for all the partition modes of the current CU, and select the partition mode with the smallest rate-distortion cost as the partition mode of the current CU; The spatiotemporally adjacent CUs include the left coding unit L-CU, the upper coding unit U-CU and the coding unit Col-CU at the same position as the current CU in the previous frame of the current CU; Col-CU, L-CU and U-CU are used to predict the depth of the current CU, and equal weights are assigned to each spatiotemporally adjacent CU.

2. The VVC coding unit fast division method based on spatiotemporal depth information according to claim 1, characterized in that: The quadtree partition depth of the maximum coding unit is represented by the depth mean of the maximum coding block: a maximum coding unit is divided into 4 CUs, the depth mean of the 4 CUs is calculated respectively, the depth values ​​of the 4 CUs are added together to obtain the average value, and the depth of the maximum coding unit is obtained.

3. The VVC coding unit fast division method based on spatiotemporal depth information according to claim 2, characterized in that: The relationship between the average depth of the quadtree partition and the current maximum coding unit is as follows: Step 1. For a maximum coding unit with optimal RDO coding, calculate the number of blocks with different depths and obtain the CU with the best depth; Step 2. Get the depth mean of three spatiotemporal adjacent blocks Col-CU, L-CU and U-CU respectively; Step 3. Take the weighted sum of the quadtree partition depth means of the coding block Col-CU of the previous frame, the coded left coding block L-CU, and the upper coding block U-CU.

Images