A low-complexity fast intra-frame encoding method for VVC

Abstract

The present invention discloses a low-complexity fast intra-frame encoding method for VVC. In the present invention, the CU partitioning module based on context correlation first partitions the CU into complex blocks and simple blocks according to the texture features of the current CU block; early termination of all types of partitioning is performed on the simple blocks, and binary tree partitioning and ternary tree partitioning are continued for the complex blocks, so as to solve the problem of excessively high complexity caused by multi-type tree partitioning. In addition, when the encoder determines the best mode of the current CU, first, the rough selection part in the intra-frame is optimized. Among the angular modes with the index values of 2 to 67 in the intra-frame angular prediction mode, a central angle is selected every 4 angles, and then 2 angular modes are added before and after in a small range based on this for search; then, the probability distribution characteristics of the most probable mode (MPM) are used to optimize the fine selection part in the intra-frame. The present invention reduces the complexity of VVC encoding while ensuring the encoding quality, and has very good encoding performance especially for high-resolution videos.

Description

Technical Field

[0001] The present invention belongs to the field of fast coding of versatile video coding (VVC), and mainly relates to a low-complexity fast intra-frame coding method for VVC. Background Art

[0002] In recent years, with the rise of new Internet applications, global traffic has continued to grow at a high speed. On the one hand, as the world enters the mobile Internet era, emerging Internet applications represented by the Internet of Things, cloud computing, and mobile Internet have rapidly emerged, prompting the continuous high-speed growth of wireless traffic and mobile device traffic. On the other hand, the further deepening of global digital transformation will continue to expand the demand for IP networks and promote the continuous increase in IP network traffic. In particular, with the rise of new applications such as 4K, 8K ultra-high-resolution videos, high-dynamic range (HDR) videos, and 360-degree videos, the existing coding standard H.265 / HEVC (High Efficiency Video Coding) can no longer meet the increasing demands of the industry. At the San Diego Conference held in the United States on April 10, 2018, the next-generation video coding standard was named H.266 / VVC (Versatile Video Coding), and the standard-setting work was officially launched. At the same time, the Joint Video Exploration Team transitioned to the Joint Video Experts Team (JVET), and the VVC reference software VTM (VVC Test Model) was launched. The goal of H.266 / VVC is to reduce the bit rate by 50% compared to H.265 / HEVC under the same quality. The intra-frame coding complexity of VVC mainly focuses on the division of coding units (CUs) and the selection of intra-frame prediction modes. There are significant differences in CU division in VVC compared to the previous generation of High Efficiency Video Coding (HEVC) standard. On the one hand, it inherits the quadtree division of HEVC. At the same time, to better adapt to the coding of ultra-high-definition videos, VVC has made significant changes to the intra-frame coding part. First, the maximum CTU size allowed in VVC is 128x128. In VVC, the concepts of PUs and TUs are removed, and the basic units of prediction and transformation are both CUs. At the same time, a multi-type tree (MTT) division method is introduced. Specifically, the CTU in VVC is first divided into different CUs according to the quaternary tree (QT). Then, the CUs at the leaf nodes of the quaternary tree can be divided according to the multi-type tree, mainly including binary tree (BT) and ternary tree (TT) divisions. Specifically, it can be further divided into the following four division types: vertical binary tree division (SPLIT_BT_VER), horizontal binary tree division (SPLIT_BT_HOR), vertical ternary tree division (SPLIT_TT_VER), and horizontal ternary tree division (SPLIT_TT_HOR). Among them, the ternary tree is divided according to the ratio of 1:2:1 of the height or width of the CU block. Figure 1They are four partitioning modes of multi-type trees. It should be noted that the leaf nodes after the MTT partitioning of CU blocks are also called CUs. The multi-type tree provides a more flexible partitioning method. Through the partitioning of MTT, each CTU can be partitioned into square blocks with equal length and width and rectangular blocks with unequal length and width. However, this makes intra-frame coding more complex because VVC needs to traverse all prediction modes and partition types of QTMT to find the best combination. Figure 2 It shows the result after the MTT partitioning of a CTU.

[0003] In VVC, the concepts of PU and TU are cancelled, so the CU is also the basic prediction unit for intra-frame and inter-frame. Intra-frame prediction means using the spatial correlation of the video frame and using the pixel values of the already encoded part in the current frame to predict the current pixel, so as to achieve the purpose of removing spatial redundancy. There are many improvements in intra-frame prediction in H.266 / VVC compared with H.265 / HEVC. In order to capture more edge directions in natural scene videos, the number of intra-frame angular prediction modes in VVC has increased from 33 in HEVC to 65. Figure 3 The dotted lines in it are the angular modes added by VVC, and these angular modes are valid for all sizes of blocks including luminance components and chrominance components. In VVC, a method combining the rough selection mode (RMD) and the most probable modes (MPM) is adopted to reduce the calculation of rate-distortion optimization (RDO) of intra-frame modes. However, the rough selection mode in VVC is further divided into two stages. First, the even angles among the angular modes from 0 to 67 are roughly selected, and then a small-range intra-frame mode search is performed on the rough selection results. In this way, the intra-frame complexity of VVC is about 34 times that of HM, so reducing the complexity of H.266 / VVC is a very important issue. Summary of the Invention

[0004] The object of the present invention is to address the disadvantage of the high coding complexity of existing VVC, and propose a low-complexity fast VVC intra-frame coding method, which reduces the coding complexity while ensuring the coding quality, especially having very good coding performance for high-resolution videos.

[0005] The present invention proposes a low-complexity fast VVC intra-frame coding method, and the specific implementation includes a CU partitioning module based on context correlation and an intra-frame mode selection module based on spatio-temporal correlation.

[0006] The CU partitioning module based on context relevance first divides the CU into complex blocks and simple blocks according to the texture characteristics of the current CU block. For simple blocks, all types of partitioning are terminated in advance. For CU blocks with complex textures, binary tree partitioning and ternary tree partitioning are continued to solve the problem of excessive complexity caused by multi-type tree partitioning. In addition, when the encoder determines the best mode of the current CU, we associate the coding context through the intra prediction mode IPM (Intra Prediction Mode) to further improve the performance of this strategy. Finally, in order to further reduce the complexity of the ternary tree partitioning in multi-type tree partitioning, the present invention proposes a CU fast decision method based on context relevance, which uses the binary tree partitioning result to make an early judgment on the ternary tree partitioning and reduces the coding complexity of the ternary tree partitioning.

[0007] The CU mode selection module based on spatio-temporal relevance first, for the rough selection part within the frame, selects a central angle every 4 angles among the angle modes with the index values of 2 to 67 in the intra-frame angle prediction mode, which can achieve a good balance between reducing the intra-frame complexity and the prediction accuracy. Then, on this basis, in order to improve the accuracy of the intra-frame prediction mode, the angle mode search is increased in a small range. Secondly, in the fine selection part, the probability of MPM being the best prediction mode is very high, and this characteristic is used to optimize the fine selection part to further reduce the time complexity.

[0008] The present invention utilizes the context relevance of CU partitioning and the spatio-temporal relevance of the intra-frame prediction mode, reduces the complexity of CU partitioning, reduces the number of selections of the intra-frame prediction mode, and effectively improves the VVC coding efficiency.

[0009] The technical solutions adopted to solve the technical problems of the present invention are as follows

[0010] (I) CU partitioning module based on context relevance

[0011] Step (I), calculate the variance Var of the CU block to judge the texture complexity of the current CU; the calculation formula of Var is as follows:

[0012]

[0013] Step (II), divide the current CU into two categories according to the threshold: Category I and Category II. Category I is the CU with simple texture, and Category II is the CU with complex texture. The specific classification rules are as follows:

[0014]

[0015] Where Var is the variance of the CU block and QP is the quantization parameter. For Category I, the partitioning of QT, BT, and TT is skipped, and for Category II, the optimization is continued.

[0016] Step (Ⅲ): For Class II, associated luminance samples are used to determine the directions of binary tree partitioning and ternary tree partitioning of the current CU block, and then the current CU partitioning rule is optimized. The rule is as follows:

[0017] 1) Calculate the standard deviations (F hor and F ver ) in the horizontal and vertical directions of CU blocks of different sizes:

[0018]

[0019]

[0020] 2) Divide the intra prediction mode IPM into two categories. The index of the horizontal mode in the IPM is 18, and the index of the vertical mode is 50. Therefore, define the IPM index numbers (11 - 25) as the horizontal direction IPM hor , and the IPM index numbers (43 - 57) as the vertical direction IPM ver . To obtain the accuracy of partitioning, the following partitioning rules are set:

[0021]

[0022]

[0023] where FH skipsplit and FV skipsplit respectively represent skipping horizontal partitioning and vertical partitioning. Judge the current block according to (3) and (4) above. When the standard deviation in the vertical direction minus the standard deviation in the horizontal direction is less than the threshold threshold i＝0,1,2 and the intra prediction mode is vertical partitioning, skip the horizontal binary tree and ternary tree partitioning. When the standard deviation in the horizontal direction minus the standard deviation in the vertical direction is less than the threshold threshold i＝0,1,2 , and the intra prediction mode is horizontal partitioning, skip the vertical binary tree and ternary tree partitioning.

[0024] Here, the selection of the threshold is explained. The VVC standard test sequences are divided into 5 categories according to different video resolutions and texture features. Select one video sequence from each category. The selected test sequences are Cactus, PartyScene, BasketballPass, FourPeople, and ChinaSpeed sequences. Here, a partitioning parameter Miss - hit is defined.

[0025]

[0026] Among them, Miss-hit represents the percentage of CUs that are not accurately partitioned. ΔBDBR and ΔTime represent the average values of the bitrate and time efficiency obtained when QP takes 22, 27, 32, and 37 respectively. We hope that the smaller Miss-hit is, the better. First, plot the curves of Miss-hit versus threshold for these 5 sequences respectively, then average the 5 curves to generate an average curve, and select the lowest point of this curve as the final threshold threshold i＝0,1,2 。

[0027] Step (Ⅳ): For CU blocks that do not meet the conditions of Step (Ⅱ) and Step (Ⅲ), perform context-based fast CU early termination, and optimize the ternary tree partition using the results of the optimal binary tree partition. The rules for CU fast partition optimization are as follows:

[0028] (1) Obtain the optimal rate-distortion costs in the horizontal and vertical directions of the binary tree partition. δ(T0) and δ(T1) represent the horizontal binary tree and vertical binary tree partitions respectively; RD(T0) and RD(T1) represent the optimal rate-distortion costs in the horizontal and vertical partition directions of the binary tree respectively.

[0029]

[0030] (2) Determine the partition direction of the optimal binary tree for the current CU. If RD(T0) < RD(T1), the current CU performs a horizontal ternary tree partition and terminates the vertical ternary tree partition judgment in advance. Correspondingly, if RD(T1) < RD(T0), the current CU performs a vertical ternary tree partition and terminates the horizontal ternary tree partition judgment in advance.

[0031] (2) Pattern selection module based on spatio-temporal correlation:

[0032] Step (1): Statistically analyze the probabilities of the optimal prediction patterns in different angular pattern sets. It is found that a central angle is selected every 4 angles among the angular patterns with index values from 2 to 34, achieving a good balance between reducing intra-frame complexity and prediction accuracy.

[0033] Step (2): Based on the central angle selected in step (1), to improve the accuracy of the intra prediction mode, add 2 angle mode searches before and after in a small range. For example, if the current mode is 50, add two angle modes to the left and right directions respectively. Combine angles 48, 49, 51, 52 and angle 50 to form an angle set. Since there are special edge angles in the selected angle set, such as angle index values 2 and 66, only one-sided angles need to be taken. That is, when the selected angle is 2, select two adjacent angles to the right, and combine angles 3, 4 and angle 2 to form an angle set. Similarly, when the selected angle is 66, select two adjacent angles to the left, and combine angles 64, 65 and angle 66 to form an angle set.

[0034] Step (3): For the refined selection stage, there is also a strong correlation in the intra prediction mode between adjacent CU blocks. The probability that MPM is used as the best prediction mode is very high. Therefore, utilize the correlation between adjacent prediction modes to optimize the mode selection module based on spatio-temporal correlation, so as to reduce the computational complexity of RDO. The main rules are as follows:

[0035] Rule 1: Denote the sum of absolute differences (Sum of Absolute Difference) of the MPM mode as SAD, and denote the cost of the sum of absolute values after Hadamard transform (SATD, Sum of Absolute Transformed Difference) as J SATD , when both SAD and J SATD are minimized simultaneously, directly use their corresponding MPM modes as the best prediction mode. Prematurely terminate the current mode selection process.

[0036] Rule 2: Screen the 6 modes in MPM. According to the minimum J calculated in the rough selection mode SATD , adaptively obtain a threshold Th,

[0037] Th = J min(SATD) × μ (9)

[0038] where, J min(SATD) is the minimum value of J of all prediction modes of the current block SATD , and μ is a proportionality coefficient. If the J SATD value of the prediction mode in MPM is less than Th, this MPM mode will not be put into the final RDO calculation, so the time complexity can be further reduced. For other cases of MPM modes, perform the original encoding operation.

[0039] The beneficial effects of the present invention are as follows:

[0040] The basic principle of the present invention is to utilize the texture features of the current frame CU and the correlation of the CUs adjacent to the current frame CU in terms of the partitioning mode and the intra-frame best prediction mode. First, for the CU blocks with simple global texture, all their partitioning modes are terminated in advance. For the remaining CU blocks, based on the texture features in the vertical and horizontal directions of the CU block, the binary tree or ternary tree partitioning is determined in advance. Then, based on the context correlation, the partitioning of the ternary tree is determined in advance using the partitioning result of the best binary tree. On the other hand, the present invention utilizes the spatio-temporal correlation of the CU block to reduce the number of angles and modes in the rough selection part and the fine selection part of the intra-frame prediction mode selection, thereby improving the coding performance of the intra-frame prediction. Through experimental measurement, on the premise of maintaining the video coding quality, the VVC fast coding method proposed by the present invention can save about 55% in coding time compared with the standard VVC coding method, while the coding bit rate only increases by about 2.5%, greatly improving the efficiency of video coding and having strong practicability. Description of the Drawings

[0041] Figure 1 Schematic diagrams of four partitioning modes of the multi-type tree for VVC;

[0042] Figure 2 Schematic diagram of the CU partitioning mode for VVC;

[0043] Figure 3 Flowchart of the method for the CU partitioning part;

[0044] Figure 4 Schematic diagram of the selection of the main direction of the intra-frame prediction mode for VVC;

[0045] Figure 5 Flowchart of the method for the CU mode selection part;

[0046] Figure 6 Curve graph of the threshold selection for the CU partitioning part;

[0047] Figure 7 Flowchart of the method of the present invention; Detailed Embodiment

[0048] The present invention will be further described below in conjunction with the drawings and embodiments.

[0049] As Figures 1-7 shown, a fast general video coding method based on context correlation adopts the VTM4.0 model of VVC video coding. The test conditions refer to the general test conditions of JVET-N1010 (JVET-N1010 CTC SDR), and the full intra-frame coding profile encoder_intra_vtm.cfg encoded using the VTM model is used.

[0050] A fast and general video coding method based on context relevance. First, the flowchart of the method for the CU partitioning part is as shown in Figure 3 Next, for the CU mode selection part, the schematic diagram of the main direction selection of the intra prediction mode is as shown in Figure 4 and the flowchart of the method is as shown in Figure 5 Among them, the specific steps of the CU partitioning module based on context relevance are as follows:

[0051] Step (I): Calculate the variance Var of the CU block to determine the texture complexity of the current CU. The calculation formula of Var is as follows:

[0052]

[0053] where W and H are the width and height of the CU, P(x, y) represents the pixel value, and (x, y) represents the position of the CU pixel.

[0054] Step (II): Classify the current CU into two categories according to the threshold: Category I and Category II. Category I is the CU with simple texture and Category II is the CU with complex texture. The specific classification rules are as follows:

[0055]

[0056] where Var is the variance of the CU block and QP is the quantization parameter. For Category I, the partitioning of QT, BT, and TT is skipped, and for Category II, the optimization continues. Among them, the threshold is determined to be 3 according to experiments.

[0057] Step (III): For Category II, associated luminance samples are used to determine the directions of the binary tree partitioning and the ternary tree partitioning of the current CU block, and then the partitioning rule of the current CU is optimized. The rule is as follows:

[0058] 1) According to the theoretical knowledge of image processing, it is known that the variance of an image can reflect the texture characteristics of the image. Therefore, for the CU block with horizontal texture, the variance value in the vertical direction is greater than that in the horizontal direction. On the contrary, for the CU block with vertical texture, the variance value in the horizontal direction is greater than that in the vertical direction. The standard deviation has the same characteristics as the variance. Calculate the standard deviations in the horizontal and vertical directions (F hor and F ver ):

[0059]

[0060]

[0061] where F hor and F verThey respectively represent the standard deviation in the horizontal direction and the standard deviation in the vertical direction of the current CU. P(x, y) represents the pixel value, and (x, y) represents the position of the CU pixel. When F hor is less than F ver , it tends to make a horizontal CU partition. When F ver is less than F hor , it tends to make a vertical CU partition. Therefore, F hor and F ver can be used to predict the directions of binary tree and ternary tree partitions, avoiding unnecessary MTT evaluations. In addition, when the coding context is selected as the best mode of the current CU, we associate the coding context through IPM (intra prediction mode) to further improve the performance of this strategy. Therefore, we divide the intra prediction mode IPM into two categories. Based on the intra prediction mode, the index of the horizontal mode is 18, and the index of the vertical mode is 50. So we define the intra prediction mode index numbers (11 - 25) as the horizontal partition direction (IPM hor ), and the intra prediction mode index numbers (43 - 57) as the vertical partition direction (IPM ver ).

[0062] To obtain the accuracy of the partition, we set a threshold threshold i＝0,1,2 here. The judgment condition for step 2 is:

[0063]

[0064]

[0065] Among them, FH skipsplit and FV skipsplit respectively represent skipping the horizontal partition and the vertical partition. According to the above (5)(6), judge the current block. When the standard deviation in the vertical direction minus the standard deviation in the horizontal direction is less than the threshold threshold i＝0,1,2 and the intra prediction mode is the vertical partition, skip the horizontal binary tree and horizontal ternary tree partitions. When the standard deviation in the horizontal direction minus the standard deviation in the vertical direction is less than the threshold threshold i＝0,1,2 and the intra prediction mode is the horizontal partition, skip the vertical binary tree and horizontal ternary tree partitions.

[0066] Among them, threshold i＝0 , threshold i＝1 , threshold i＝2 are determined to be 70, 90, 30 according to experiments. The specific calculation method of the threshold threshold i＝0,1,2 is as follows:

[0067] The VVC standard test sequences are divided into 5 categories according to different video resolutions and texture features. One video sequence is selected from each category. The selected test sequences are Cactus, PartyScene, BasketballPass, FourPeople, and ChinaSpeed sequences. Here, a partitioning parameter Miss-hit is defined.

[0068]

[0069] Among them, Miss-hit represents the percentage of CUs that are not accurately partitioned. ΔBDBR and ΔTime represent the percentage of bitrate savings and time savings respectively. Both ΔBDBR and ΔTime are the average values of the bitrate and time efficiency obtained when QP takes 22, 27, 32, and 37. We hope that the smaller the Miss-hit, the better. The experimental results show that the value of Miss-hit is very small. The schematic diagram of the threshold selection for CU partitioning is as Figure 6 shown. First, draw the curves of Miss-hit vs. threshold for these 5 sequences respectively, then average the 5 curves to generate an average curve, and select the lowest point of this curve as the final threshold threshold i＝0,1,2 .

[0070] Step (Ⅳ): For CU blocks that do not meet the conditions of Step (Ⅱ) and Step (Ⅲ), a context-based fast CU early termination algorithm is proposed to optimize the ternary tree partitioning using the results of the optimal binary tree partitioning. The rules for CU fast partitioning optimization are as follows:

[0071] (1) Obtain the optimal rate-distortion costs in the horizontal and vertical directions of the binary tree partitioning. δ(T0) and δ(T1) represent the horizontal binary tree and vertical binary tree partitions respectively. In addition, RD(T0) and RD(T1) represent the optimal rate-distortion costs in the horizontal and vertical partitioning directions of the binary tree respectively.

[0072]

[0073] (2) Determine the partitioning direction of the optimal binary tree for the current CU. If RD(T0) < RD(T1), the current CU performs horizontal ternary tree partitioning and terminates the vertical ternary tree partitioning judgment in advance. Correspondingly, if RD(T1) < RD(T0), the current CU performs vertical ternary tree partitioning and terminates the horizontal ternary tree partitioning judgment in advance.

[0074] The specific steps of the mode selection module based on spatio-temporal correlation are as follows:

[0075] Step (1): In the rough selection stage, in each angular mode with index values from 2 to 34, a central angle is selected every 4 angles. After the initial screening, the selected modes are as follows: 2, 6, 8, 10, 14, 18, 22, 26, 30, 34, 38, 42, 46, 50, 54, 58, 62, 66 are the central angle modes. Calculate the J SATD value of these modes and sort them in descending order. Select the mode with the smallest J SATD as IP mode .

[0076] Step (2): Based on the central angles selected in step (1) above, in order to improve the accuracy of the intra prediction mode, search for 2 additional angular modes in a small range before and after. For example, if the current mode is 50, add two angular modes to the left and right directions respectively. Combine angles 48, 49, 51, 52 with angle 50 to form an angle set. Since there are special edge angles in the selected angle set, such as angle index values 2 and 66, only one-sided angles need to be taken. That is, when the selected angle is 2, select two adjacent angles to the right, and combine angles 3, 4 with angle 2 to form an angle set. Similarly, when the selected angle is 66, select two adjacent angles to the left, and combine angles 64, 65 with angle 66 to form an angle set.

[0077] Step (3): Based on steps (1) and (2) above, in order to increase the accuracy of intra mode selection, add two non-angular modes, the Planar mode and the DC mode, to the final rough selection part set.

[0078] Step (4): In the fine selection stage, since the MPM mode has a very high probability of being the best prediction mode, it can be used to reduce the computational amount of RDO. The main specific rules are as follows:

[0079] Rule 1: Denote the sum of absolute differences (SAD) of the MPM mode as SAD, and denote the cost of the sum of absolute transformed differences (SATD) after Hadamard transform as J SATD . When both SAD and J SATD are the smallest, directly use their corresponding MPM modes as the best prediction mode. Terminate the current mode selection process in advance.

[0080] Rule 2: Screen the 6 modes in the MPM. Based on the smallest J SATD calculated from the rough selection mode, adaptively obtain a threshold Th.

[0081] Th=J min(SATD) ×μ (9)

[0082] Among them, J min(SATD) is the J of all prediction modes of the current block SATD The minimum value of the value, μ is the proportional coefficient. If the J of the prediction model in MPM SATD If the value is less than Th, the MPM mode will not be included in the final RDO calculation, so the time complexity can be further reduced. For the MPM mode in other cases, the original encoding operation is performed. According to the experiment, the threshold μ=1.3 is determined.

[0083] at last, Figure 7 The method includes a CU partitioning module based on context correlation and a CU mode selection module based on spatiotemporal correlation.

Claims

1. A low-complexity fast intra-frame coding method for VVC, characterized in that It includes a CU partitioning module based on context relevance and an intra-mode selection module based on spatio-temporal correlation; The CU partitioning module based on context relevance first divides the CU into a complex block and a simple block according to the texture feature of the current CU block; for the simple block, all types of partitioning are terminated in advance, and for the complex block, binary tree partitioning and ternary tree partitioning are continued; at the same time, when the encoder determines the best mode of the current CU: the CU mode selection module based on spatio-temporal correlation first selects a central angle every 4 angles among the angle modes with the index value of 2 to 67 in the intra-frame rough selection part; then based on this, 2 angle modes are added before and after in a small range for search; secondly, in the fine selection part, the probability that the MPM is the best prediction mode is very high, and this feature is used to optimize the fine selection part; The CU partitioning module based on context correlation is specifically implemented as follows: Step (I), calculate the variance Var of the CU block to judge the texture complexity of the current CU; the calculation formula of Var is as follows: where W and H respectively represent the width and height of the current CU; P(x, y) represents the pixel value at the position (x, y) in the CU; Step (II), divide the current CU into two categories according to the threshold: Category I and Category II; Category I is the CU with simple texture, and Category II is the CU with complex texture. The specific classification rules are as follows: where Var is the variance of the CU block, is the classification threshold determined by experiments, and QP is the quantization parameter; for type I, the partitioning of QT, BT, and TT is skipped, and for type II, the optimization continues; Step (Ⅲ), for Category II, associated luminance samples are used to determine the directions of binary tree partitioning and ternary tree partitioning of the current CU block, and then the current CU partitioning rule is optimized; Step (Ⅳ), for the CU block that does not meet the conditions of Step (Ⅱ) and Step (Ⅲ), context-based fast CU early termination is performed, and the result of the best binary tree partitioning is used to optimize the ternary tree partitioning.

2. The low-complexity fast intra-frame encoding method of VVC according to claim 1, wherein The rule described in Step (Ⅲ) is as follows: 1) Calculate the standard deviations F in both the horizontal and vertical directions for CU blocks of different sizes hor and F ver : 2) The intra prediction mode (IPM) is divided into two categories. The index of the horizontal mode in the IPM is 18, and the index of the vertical mode is 50. Therefore, the IPM index numbers 11 to 25 are defined as the horizontal direction IPM hor , and the IPM index numbers 43 to 57 are defined as the vertical direction IPM ver ; To obtain the accuracy of the division, the following division rules are set: Among them, FH skipsplit and FV skipsplit respectively represent skipping horizontal partitioning and vertical partitioning. According to the above (3) and (4), the current block is judged. When the standard deviation in the vertical direction minus the standard deviation in the horizontal direction is less than the threshold threshold i＝0,1,2 and the intra prediction mode is vertical partitioning, the horizontal binary tree and ternary tree partitions are skipped. When the standard deviation in the horizontal direction minus the standard deviation in the vertical direction is less than the threshold threshold i＝0,1,2 , and the intra prediction mode is horizontal partitioning, the vertical binary tree and ternary tree partitions are skipped.

3. According to the low-complexity fast VVC intra-frame encoding method described in claim 2, the selection of the threshold is implemented as follows: The VVC standard test sequences are divided into 5 categories according to different video resolutions and texture features, and a video sequence is selected from each category. The selected test sequences are Cactus, PartyScene, BasketballPass, FourPeople, and ChinaSpeed sequences. Here, a partitioning parameter Miss-hit is defined, Among them, Miss-hit represents the percentage of inaccurate CU partitioning; ΔBDBR and ΔTime represent the average values of bitrate and time efficiency obtained when QP takes 22, 27, 32, and 37, respectively. First, plot the curves of Miss-hit versus threshold for these 5 sequences separately, then average the 5 curves to generate an average curve, and select the lowest point of this curve as the final threshold threshold i＝0,1,2 。 4. A low-complexity fast intra-frame coding method for VVC according to claim 2, characterized in that The rule for CU fast partitioning optimization in Step (Ⅳ) is: (1), obtain the best rate-distortion costs in the horizontal and vertical directions of binary tree partitioning. δ(T0) and δ(T1) respectively represent the horizontal binary tree and vertical binary tree partitioning; RD(T0) and RD(T1) respectively represent the best rate-distortion costs in the horizontal and vertical partitioning directions of the binary tree; (2) Determine the partitioning direction of the current CU's optimal binary tree. If RD(T0) < RD(T1), the current CU performs horizontal ternary tree partitioning and terminates the vertical ternary tree partitioning judgment in advance; correspondingly, if RD(T1) < RD(T0), the current CU performs vertical ternary tree partitioning and terminates the horizontal ternary tree partitioning judgment in advance.

5. A low-complexity fast intra-frame VVC encoding method according to claim 1 or 4, characterized in that it is based on a spatio-temporal correlation-based mode selection module, and is specifically implemented as follows: Step (1): Statistically analyze the probabilities of the best prediction modes in different angular mode sets, and select a central angle every 4 angles among the angular modes with index values from 2 to 34. Step (2): Based on the central angles selected in step (1), search for 2 additional angular modes before and after in a small range: If the current mode is 50, add two angular modes to the left and right directions respectively, and form an angular set by combining angles 48, 49, 51, 52 and angle 50; if there are special edge angles in the selected angular set, only take the angles on one side, that is, when the angle index value is 2, select two adjacent angles to the right, and form an angular set by combining angles 3, 4 and angle 2; similarly, when the selected angle is 66, select two adjacent angles to the left, and form an angular set by combining angles 64, 65 and angle 66. Step (3): For the refined selection stage, there is also a strong correlation between the intra-frame prediction modes of adjacent CU blocks, and the probability of MPM being the best prediction mode is very high. Therefore, the spatio-temporal correlation-based mode selection module is optimized by using the correlation between adjacent prediction modes; the main rules are as follows: Rule 1: Denote the sum of absolute errors of the MPM mode as SAD, and denote the cost of the sum of absolute values after Hadamard transform as J SATD , when both SAD and J SATD are minimized simultaneously, directly use their corresponding MPM modes as the best prediction modes; terminate the current mode selection process prematurely; Rule 2: Screen the 6 patterns in the MPM, and obtain a threshold Th adaptively according to the minimum J calculated in the roughly selected patterns SATD , and adaptively obtain a threshold Th Th = J min(SATD) × μ (9) Among them, J min(SATD) is the minimum value of J for all prediction modes of the current block, where μ is a scaling factor; if the J value of the prediction mode in MPM is less than Th, this MPM mode will not be put into the final RDO calculation; for other MPM modes in other cases, the original encoding operation is performed. SATD value is less than Th, then this MPM mode will not be put into the final RDO calculation; for other MPM modes in other cases, the original encoding operation is performed. SATD value is less than Th, then this MPM mode will not be put into the final RDO calculation; for other MPM modes in other cases, the original encoding operation is performed.

Images