VVC fast intra-frame coding mode decision-making method based on machine learning

Abstract

The invention relates to a VVC fast intra-frame coding mode decision-making method based on machine learning, which belongs to the field of video coding and comprises the following steps of: acquiring texture information, sub-block texture information and field information of a current to-be-coded CU (Control Unit) for LightGBM (LightGBM) model prediction; the method comprises the following steps: firstly, predicting probabilities of three categories by using a trained LightGBM model, wherein the three categories are respectively 0 (angle prediction mode), 1 (ISP mode) and 2 (MIP mode); comparing the probability with a threshold value, and removing the category of which the probability is lower than the threshold value, so as to reduce the number of candidate modes of which the complete rate distortion optimization process needs to be performed; in addition, under the condition that the category 0 is reserved, the probability of 67 angle prediction modes is further predicted by using another LightGBM model, and the prediction mode with lower probability is removed. According to the method, on the premise of ensuring the video quality, the intra-frame coding time of H.266 / VVC can be remarkably saved, and the method is suitable for video coding application scenes with high real-time performance requirements.

Description

Technical Field

[0001] The present invention belongs to the field of video coding, and more specifically, to a fast intra-frame coding mode decision method suitable for Versatile Video Coding (VVC), which can be applied to video coding scenarios with high real-time requirements. Background Art

[0002] With the popularization of Internet applications, users' expectations for videos are constantly increasing. They hope to get a more realistic and clear visual experience, and expect the smooth performance of video playback to reach a higher level. In order to meet the needs of users, various emerging video applications have emerged, such as Ultra High Definition (UHD), High Frame Rate (HFR), High Dynamic Range (HDR), Wide Color Gamut (WCG), Virtual Reality (VR) and Augmented Reality (AR). However, the surge in video data has put tremendous pressure on the limited communication transmission bandwidth and storage resources. Therefore, the Joint Video Experts Team (JVET) has developed the next-generation video coding standard VVC (Versatile Video Coding). Compared with its predecessor, High Efficiency Video Coding (HEVC), VVC uses a variety of new coding technologies, such as the Coding Unit (CU) division structure based on Quad-tree Plus Multi-typeTree (QTMT), more angle prediction modes (from 35 in HEVC to 67 in VVC), Wide Angle Intra Prediction (WAIP), Position Dependent Intra Prediction Combination (PDPC), Multiple Reference Line (MRL) prediction, Intra Sub-Partitions Mode (ISP) and Matrix Weighted Intra Prediction (MIP) etc. These new coding technologies have improved the coding efficiency of VVC by nearly 50%, but at the same time, the coding complexity has also increased significantly. There are two main reasons for the substantial increase in coding complexity: the QTMT-based CU partition structure and the intra-coding mode search process based on rate distortion optimization (RDO). The present invention focuses on reducing the complexity of intra-coding mode search.The RDO-based intra-frame coding mode search process performs a complete RDO process (including prediction, transformation, quantization and entropy coding processes) on each mode in the candidate mode list (CML) and calculates the rate-distortion cost of each mode. The mode with the lowest rate-distortion cost is selected as the optimal intra-frame coding mode. This brute force search process greatly increases the coding complexity. Therefore, the present invention reduces the coding complexity by reducing the number of modes in the candidate mode list.

[0003] The proposed method LightGBM model predicts the intra-frame coding mode, and uses the prediction results to remove redundant intra-frame coding modes to reduce the computational complexity of the RDO process and significantly reduce the coding time of VVC intra-frame coding. First, the original encoder is used to encode the video sequence, and the feature data and intra-frame coding mode are recorded in the process, and the corresponding data set is established for training the Light Gradient Boosting Machine (LightGBM) model; secondly, the hyperparameters of the LightGBM model training are determined, and the data set training model is loaded; finally, the encoder loading model is used to predict the CU intra-frame coding mode, and the prediction results are used to remove the redundant RDO process. The present invention can be used for coding under the VVC intra-frame coding configuration, and significantly reduces the coding time of VVC while ensuring that the coding efficiency loss is minimal. Summary of the invention

[0004] The present invention aims to solve the above problems of the prior art. A VVC fast intra-frame coding mode decision method based on machine learning is proposed. The technical solution of the present invention is as follows:

[0005] A VVC fast intra-frame coding mode decision method based on machine learning comprises the following steps:

[0006] S1. Select CU features for training Lightweight Gradient Boosting Machine LightGBM model 1, which implements three categories of probability prediction, namely angle prediction mode, intra sub-block partitioning (ISP) mode, and matrix-based prediction (MIP) mode; CU features include two categories of features, namely texture information and sub-block texture information;

[0007] S2. Select CU features for training LightGBM model 2, which implements probability prediction of 67 prediction modes (including DC, plane and 65 different angles) in angle prediction mode; CU features include 3 categories of features, namely texture information, sub-block texture information and context information;

[0008] S3. Use the modified VVC test reference model encoder to encode the video sequence, collect the texture and intra-frame coding mode data of the CU, establish a training data set, and train LightGBM model 1 and LightGBM model 2;

[0009] S4. Use the LightGBM model 1 and LightGBM model 2 trained in step S3 to predict the CU intra-frame coding mode, skip the redundant CU coding mode rate distortion optimization (RDO) process, and realize the fast intra-frame coding mode decision of CU.

[0010] Further, in step S1, the texture information includes horizontal and vertical gradients, gradient ratio, variance, entropy and kurtosis of the CU;

[0011] The horizontal and vertical gradients are calculated using the Sobel operator, and the calculation formula is as follows:

[0012]

[0013] G in formula (1) h (x,y) and G v (x, y) represents the horizontal gradient and vertical gradient of the pixel at the coordinate (x, y) respectively; A(x, y) represents the pixel matrix centered at the coordinate (x, y); then, the average gradient of the current CU pixel in the horizontal and vertical directions is calculated using formula (2);

[0014]

[0015] In formula (2), |G h (x,y)| and|G v (x,y)|respectively the absolute value of the horizontal gradient and vertical gradient of the pixel at the (x,y) coordinate; Gradient x and Gradient y They are the horizontal average gradient and vertical average gradient of the current CU respectively;

[0016] The ratio of horizontal and vertical gradients is calculated as follows:

[0017]

[0018] Among them, GradientRatio represents the ratio of the horizontal gradient to the vertical gradient of the current CU;

[0019] The variance calculation formula is as follows:

[0020]

[0021] Where P(x,y) is the pixel value of the brightness component of the CU at the (x,y) coordinate; W and H represent the width and height of the CU respectively; Mean is the average pixel value of the CU; Variance is the pixel variance of the CU;

[0022] The calculation formula of entropy is as follows:

[0023]

[0024] Where L is the total number of gray levels in the brightness component of the CU; P(i) is the normalized probability of gray level i; Entropy is the entropy of the CU image;

[0025] The calculation formula for kurtosis is as follows:

[0026]

[0027] Where N is the total number of pixels in the CU; L is the total number of gray levels; x i is the number of pixels of gray level i; μ is the gray mean of CU; σ is the gray standard deviation of CU; Kurtosis is the kurtosis value of CU.

[0028] Furthermore, in step S1, the sub-block texture information includes: sub-block variance and variance of sub-block variance; the sub-block variance is obtained by dividing the current CU to be encoded into 4 blocks horizontally and vertically, and calculating the variance of each sub-block using formula (4). The calculation formula of the variance of sub-block variance is as follows:

[0029]

[0030] Among them, Variance i Mean represents the variance of the i-th sub-block after the CU is divided horizontally / vertically into 4 blocks; sub-block Variance represents the mean of the variance of the sub-block; sub-block represents the variance of the sub-block variance.

[0031] Furthermore, in step S2, the texture information and sub-block texture information are exactly the same as those in step S1, and the context information includes: the mean value of the upper and left reference row pixels, and the calculation formula is as follows:

[0032]

[0033] Where N is the total number of pixels in the reference row; P i Represents the i-th pixel value; Mean left Represents the pixel mean of the left reference row of the current CU; Mean top Represents the pixel mean of the upper reference row of the current CU.

[0034] Furthermore, in step S3, the encoded data of the collected CU is obtained by encoding the picture or video sequence using the modified VTM encoder, and the feature data and intra-frame coding mode of the CU are recorded in the process; wherein the picture sequence used to collect samples includes 6 resolutions, namely: 416×240, 832×480, 1280×720, 1920×1080, 2560×1600 and 3840×2160.

[0035] Furthermore, in step S3, the training parameters of the LightGBM model are set to: num_leaves=31, learning_rate=0.1, early_stop_rounds=5; wherein num_leaves is the maximum number of leaves for each subtree, which is used to control the complexity of the model. A higher value can improve the fitting ability of the model, and a lower value can reduce the overfitting of the model; learning_rate is the learning rate of model training, which is used to control the learning speed of the model; early_stop_rounds is the number of early stopping rounds of model training. When the loss of the model does not change during the specified number of rounds of training, the training process of the model is stopped.

[0036] Furthermore, in step S4, the trained LightGBM model is used to predict the CU intra-frame coding mode, select the optimal CU intra-frame coding mode or skip the redundant CU intra-frame coding mode, and implement the specific process of fast intra-frame coding mode decision of CU as follows:

[0037] S41, obtaining a candidate mode list through a rough screening process of the original encoder, and proceeding to step S42;

[0038] S42, check whether the current CU to be encoded is a luminance component, and whether the CU size meets the following size requirements: 32×32, 32×16, 16×32, 32×8, 8×32, 32×16, 16×32, 16×16, 16×8, 8×16, 8×8; if so, proceed to step S43; otherwise, proceed to step S45;

[0039] S43, extract the texture information and sub-block texture information of the current CU as features, and use LightGBM model 1 to predict the probabilities of the three categories of angle mode, ISP mode and MIP mode; compare the probabilities of the three categories with the thresholds respectively, if they are less than the thresholds, remove the corresponding category of the mode in the candidate mode list; otherwise, retain the corresponding category of the mode and enter step S44;

[0040] S44, if the probability of the angle mode category is greater than the threshold, the texture information, sub-block texture information and context texture information of the current CU are extracted as features, and the LightGBM model 2 is used to predict the probabilities of 67 angle prediction modes; and the angle modes in the candidate mode list obtained in step S41 are compared with the probabilities, and only the modes in the top-3 are retained, and step S45 is entered;

[0041] S45, using the complete rate-distortion optimization process of the original encoder to calculate the rate-distortion cost of each mode in the candidate mode list, and selecting a mode with the lowest rate-distortion cost as the optimal intra-frame coding mode, and proceeding to step S46;

[0042] S46. End the intra-frame coding mode search process of the current CU.

[0043] A storage medium stores a computer program therein, wherein when the computer program is read by a processor, any one of the above-mentioned VVC intra-frame fast coding mode decision methods based on machine learning is executed.

[0044] The advantages and beneficial effects of the present invention are as follows:

[0045] Aiming at the problem that the intra-frame coding complexity of the video encoder implemented by the VVC standard is too high and the coding time is too long, the present invention proposes a VVC fast intra-frame coding mode decision method based on machine learning. First, the original encoder is used to encode the video sequence, and the feature data and intra-frame coding mode are recorded in the process, and the corresponding data set is established for training the Light Gradient Boosting Machine (LightGBM) model; secondly, the hyperparameters of the LightGBM model training are determined, and the data set training model is loaded; finally, the encoder is used to load the model to predict the CU intra-frame coding mode, and the prediction result is used to remove the redundant RDO process. Through the above steps, VVC fast intra-frame coding mode decision can be achieved.

[0046] The innovation of the present invention mainly lies in claim 1 and claim 6. The method of claim 1 is used to complete the processes of feature design, data set construction, model training and candidate pattern list clipping. The method of claim 6 uses two trained LightGBM models to predict the CU intra-frame mode, and uses the prediction results to remove the intra-frame coding mode with lower probability, which can effectively simplify the CU intra-frame coding process and reduce the coding complexity. The present invention makes full use of the texture information of the CU as a feature to train the LightGBM model, and effectively uses two LightGBM models to complete different prediction tasks, which can more accurately remove redundant intra-frame coding modes, while ensuring the quality of video coding, and significantly reduces the coding time. Therefore, the method of the present invention has strong innovation and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The present invention provides a flowchart of a VVC fast intra-frame coding mode decision method based on machine learning in a preferred embodiment;

[0048] Figure 2 It is a VVC intra-frame coding mode decision flow chart. DETAILED DESCRIPTION

[0049] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.

[0050] The technical solution of the present invention to solve the above technical problems is:

[0051] Figure 1 The flowchart of a VVC fast intra-frame coding mode decision method based on machine learning in the present invention is as follows:

[0052] S1. Select CU features for training LightGBM model 1. The model implements probability prediction of three categories, namely angle prediction mode, intra sub-block partitioning (ISP) mode, and matrix-based prediction (MIP) mode; CU features include two categories of features, namely texture information and sub-block texture information;

[0053] S2. Select CU features for training LightGBM model 2, which implements probability prediction of 67 prediction modes (including DC, plane and 65 different angles) in angle prediction mode; CU features include 3 categories of features, namely texture information, sub-block texture information and context information;

[0054] S3. Use the modified VVC test reference model encoder to encode the video sequence, collect data such as CU texture and intra-frame coding mode, establish a training data set, and train two LightGBM models;

[0055] S4. Use the LightGBM model trained in step S3 to predict the CU intra-frame coding mode, skip the redundant CU coding mode rate distortion optimization (RDO) process, and realize the fast intra-frame coding mode decision of CU.

[0056] 2. A VVC fast intra-frame coding mode decision method based on machine learning according to claim 1, characterized in that in step S1, the texture information includes horizontal and vertical gradients, gradient ratios, variance, entropy and kurtosis of the CU;

[0057] The horizontal and vertical gradients are calculated using the Sobel operator, and the calculation formula is as follows:

[0058]

[0059] G in formula (1) h (x,y) and G v (x, y) represents the horizontal gradient and vertical gradient of the pixel at the coordinate (x, y) respectively; A(x, y) represents the pixel matrix centered at the coordinate (x, y); then, the average gradient of the current CU pixel in the horizontal and vertical directions is calculated using formula (2);

[0060]

[0061] In formula (2), |G h (x,y)| and|G v (x,y)|respectively the absolute value of the horizontal gradient and vertical gradient of the pixel at the (x,y) coordinate; Gradient x and Gradient y They are the horizontal average gradient and vertical average gradient of the current CU respectively;

[0062] The ratio of horizontal and vertical gradients is calculated as follows:

[0063]

[0064] Among them, GradientRatio represents the ratio of the horizontal gradient to the vertical gradient of the current CU;

[0065] The variance calculation formula is as follows:

[0066]

[0067] Where P(x,y) is the pixel value of the brightness component of the CU at the (x,y) coordinate; W and H represent the width and height of the CU respectively; Mean is the average pixel value of the CU; Variance is the pixel variance of the CU;

[0068] The calculation formula of entropy is as follows:

[0069]

[0070] Where L is the total number of gray levels in the brightness component of the CU; P(i) is the normalized probability of gray level i; Entropy is the entropy of the CU image;

[0071] The calculation formula for kurtosis is as follows:

[0072]

[0073] Where N is the total number of pixels in the CU; L is the total number of gray levels; x i is the number of pixels of gray level i; μ is the gray mean of CU; σ is the gray standard deviation of CU; Kurtosis is the kurtosis value of CU.

[0074] 3. The VVC fast intra-frame coding mode decision method based on machine learning according to claim 1, characterized in that in the step S1, the sub-block texture information includes: sub-block variance and variance of sub-block variance; the sub-block variance is obtained by dividing the current CU to be encoded into 4 blocks horizontally and vertically, and calculating the variance of each sub-block respectively using formula (4), and the calculation formula of the variance of the sub-block variance is as follows:

[0075]

[0076] Among them, Variance i Mean represents the variance of the i-th sub-block after the CU is divided horizontally / vertically into 4 blocks; sub-block Variance represents the mean of the variance of the sub-block; sub-block represents the variance of the sub-block variance.

[0077] 4. A VVC fast intra-frame coding mode decision method based on machine learning according to claim 1, characterized in that in step S2, the texture information and the sub-block texture information are exactly the same as those in step S1, and the context information includes: the mean of the upper and left reference row pixels, and the calculation formula is as follows:

[0078]

[0079] Where N is the total number of pixels in the reference row; P iRepresents the i-th pixel value; Mean left Represents the pixel mean of the left reference row of the current CU; Mean top Represents the pixel mean of the upper reference row of the current CU.

[0080] 5. A VVC fast intra-frame coding mode decision method based on machine learning according to claim 1, characterized in that in the step S3, the coded data of the collected CU is obtained by encoding the picture or video sequence using the modified VTM encoder, and the feature data and intra-frame coding mode of the CU are recorded in the process; wherein the picture sequence used for collecting samples includes 6 resolutions, namely: 416×240, 832×480, 1280×720, 1920×1080, 2560×1600 and 3840×2160;

[0081] In the step S3, the training parameters of the LightGBM model are set as: num_leaves=31, learning_rate=0.1, early_stop_rounds=5; wherein num_leaves is the maximum number of leaves for each subtree, which is used to control the complexity of the model. A higher value can improve the fitting ability of the model, and a lower value can reduce the overfitting of the model; learning_rate is the learning rate of model training, which is used to control the learning speed of the model; early_stop_rounds is the number of early stop rounds of model training. When the loss of the model does not change during the specified number of rounds of training, the training process of the model is stopped.

[0082] 6. According to a VVC fast intra-frame coding mode decision method based on machine learning according to claim 1, it is characterized in that in the step S4, the trained LightGBM model is used to predict the CU intra-frame coding mode, select the optimal CU intra-frame coding mode or skip the redundant CU intra-frame coding mode, and the specific process of implementing the CU fast intra-frame coding mode decision is as follows:

[0083] S41, obtaining a candidate mode list through a rough screening process of the original encoder, and proceeding to step S42;

[0084] S42, check whether the current CU to be encoded is a luminance component, and whether the CU size meets the following size requirements: 32×32, 32×16, 16×32, 32×8, 8×32, 32×16, 16×32, 16×16, 16×8, 8×16, 8×8; if so, proceed to step S43; otherwise, proceed to step S45;

[0085] S43, extract the texture information and sub-block texture information of the current CU as features, and use LightGBM model 1 to predict the probabilities of the three categories of angle mode, ISP mode and MIP mode; compare the probabilities of the three categories with the thresholds respectively, if they are less than the thresholds, remove the corresponding category of the mode in the candidate mode list; otherwise, retain the corresponding category of the mode and enter step S44;

[0086] S44, if the probability of the angle mode category is greater than the threshold, the texture information, sub-block texture information and context texture information of the current CU are extracted as features, and the LightGBM model 2 is used to predict the probabilities of 67 angle prediction modes; and the angle modes in the candidate mode list obtained in step S41 are compared with the probabilities, and only the modes in the top-3 are retained, and step S45 is entered;

[0087] S45, using the complete rate-distortion optimization process of the original encoder to calculate the rate-distortion cost of each mode in the candidate mode list, and selecting a mode with the lowest rate-distortion cost as the optimal intra-frame coding mode, and proceeding to step S46;

[0088] S46. End the intra-frame coding mode search process of the current CU.

[0089] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions.

[0090] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0091] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0092] The above embodiments should be understood to be only used to illustrate the present invention and not to limit the protection scope of the present invention. After reading the contents of the present invention, technicians can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A VVC fast intra-frame coding mode decision method based on machine learning, characterized in that: The following steps are involved: S1, select CU features for training Lightweight Gradient Boosting Machine LightGBM model 1, which implements three categories of probability prediction, namely angle prediction mode, intra-frame sub-block partitioning ISP mode, and matrix-based prediction MIP mode; CU features include two categories of features, namely texture information and sub-block texture information; S2. Select CU features for training LightGBM model 2, which implements probability prediction of 67 prediction methods in angle prediction mode; the 67 prediction methods include DC, plane and 65 different angles, and the CU features include 3 categories of features, namely texture information, sub-block texture information and context information; S3. Use the modified VVC test reference model encoder to encode the video sequence, collect the texture and intra-frame coding mode data of the CU, establish a training data set, and train LightGBM model 1 and LightGBM model 2; S4. Use the LightGBM model 1 and LightGBM model 2 trained in step S3 to predict the CU intra-frame coding mode, skip the redundant CU coding mode rate-distortion optimization RDO process, and realize the fast intra-frame coding mode decision of CU.

2. The VVC fast intra-frame coding mode decision method based on machine learning according to claim 1, characterized in that: In step S1, the texture information includes horizontal and vertical gradients, gradient ratio, variance, entropy and kurtosis of the CU; The horizontal and vertical gradients are calculated using the Sobel operator, and the calculation formula is as follows: G in formula (1) h (x,y) and G v (x, y) represents the horizontal gradient and vertical gradient of the pixel at the coordinate (x, y) respectively; A(x, y) represents the pixel matrix centered at the coordinate (x, y); then, the average gradient of the current CU pixel in the horizontal and vertical directions is calculated using formula (2); In formula (2), |G h (x,y)| and|G v (x,y)|respectively the absolute value of the horizontal gradient and vertical gradient of the pixel at the (x,y) coordinate; Gradient x and Gradient y They are the horizontal average gradient and vertical average gradient of the current CU respectively; The ratio of horizontal and vertical gradients is calculated as follows: Among them, GradientRatio represents the ratio of the horizontal gradient to the vertical gradient of the current CU; The variance calculation formula is as follows: Where P(x,y) is the pixel value of the brightness component of the CU at the (x,y) coordinate; W and H represent the width and height of the CU respectively; Mean is the average pixel value of the CU; Variance is the pixel variance of the CU; The calculation formula of entropy is as follows: Where L is the total number of gray levels in the brightness component of the CU; P(i) is the normalized probability of gray level i; Entropy is the entropy of the CU image; The calculation formula for kurtosis is as follows: Where N is the total number of pixels in the CU; L is the total number of gray levels; x i is the number of pixels of gray level i; μ is the gray mean of CU; σ is the gray standard deviation of CU; Kurtosis is the kurtosis value of CU.

3. The VVC fast intra-frame coding mode decision method based on machine learning according to claim 1, characterized in that: In step S1, the sub-block texture information includes: sub-block variance and variance of sub-block variance; the sub-block variance is obtained by dividing the current CU to be encoded into 4 blocks horizontally and vertically, and calculating the variance of each sub-block using formula (4). The calculation formula of the variance of sub-block variance is as follows: Among them, Variance i Mean represents the variance of the i-th sub-block after the CU is divided horizontally / vertically into 4 blocks; sub-block Variance represents the mean of the variance of the sub-block; sub-block represents the variance of the sub-block variance.

4. The VVC fast intra-frame coding mode decision method based on machine learning according to claim 1, characterized in that: In step S2, the texture information and sub-block texture information are exactly the same as those in step S1, and the context information includes: the mean value of the upper and left reference row pixels, and the calculation formula is as follows: Where N is the total number of pixels in the reference row; P i Represents the i-th pixel value; Mean left Represents the pixel mean of the left reference row of the current CU; Mean top Indicates the pixel mean of the upper reference row of the current CU.

5. The VVC fast intra-frame coding mode decision method based on machine learning according to claim 1, characterized in that: In the step S3, the encoded data of the CU is obtained by encoding the picture or video sequence using the modified VTM encoder, and the feature data and intra-frame coding mode of the CU are recorded in the process; wherein the picture sequence used for collecting samples includes 6 resolutions, namely: 416×240, 832×480, 1280×720, 1920×1080, 2560×1600 and 3840×2160.

6. The VVC fast intra-frame coding mode decision method based on machine learning according to claim 5, characterized in that: In the step S3, the training parameters of the LightGBM model are set as: num_leaves=31, learning_rate=0.1, early_stop_rounds=5; wherein num_leaves is the maximum number of leaves for each subtree, which is used to control the complexity of the model. A higher value can improve the fitting ability of the model, and a lower value can reduce the overfitting of the model; learning_rate is the learning rate of model training, which is used to control the learning speed of the model; early_stop_rounds is the number of early stop rounds of model training. When the loss of the model does not change during the specified number of rounds of training, the training process of the model is stopped.

7. The VVC fast intra-frame coding mode decision method based on machine learning according to claim 1, characterized in that: In step S4, the trained LightGBM model is used to predict the CU intra-frame coding mode, select the optimal CU intra-frame coding mode or skip the redundant CU intra-frame coding mode, and implement the specific process of fast intra-frame coding mode decision of CU as follows: S41, obtaining a candidate mode list through a rough screening process of the original encoder, and proceeding to step S42; S42, check whether the current CU to be encoded is a luminance component, and whether the CU size meets the following size requirements: 32×32, 32×16, 16×32, 32×8, 8×32, 32×16, 16×32, 16×16, 16×8, 8×16, 8×8; if so, proceed to step S43; otherwise, proceed to step S45; S43, extracting the texture information of the current CU and the sub-block texture information as features, and using the LightGBM model 1 to predict the probabilities of the three categories of angle mode, ISP mode and MIP mode; The probabilities of the three categories are compared with the thresholds respectively. If the probabilities are less than the thresholds, the patterns of the corresponding categories in the candidate pattern list are removed; otherwise, the patterns of the corresponding categories are retained and the process proceeds to step S44; S44, if the probability of the angle mode category is greater than the threshold, the texture information, sub-block texture information and context texture information of the current CU are extracted as features, and the LightGBM model 2 is used to predict the probabilities of 67 angle prediction modes; and the angle modes in the candidate mode list obtained in step S41 are compared with the probabilities, and only the modes in the top-3 are retained, and step S45 is entered; S45, using the complete rate-distortion optimization process of the original encoder to calculate the rate-distortion cost of each mode in the candidate mode list, and selecting a mode with the lowest rate-distortion cost as the optimal intra-frame coding mode, and proceeding to step S46; S46. End the intra-frame coding mode search process of the current CU.

8. A storage medium storing a computer program, characterized in that: When the computer program is read by the processor, the VVC intra-frame fast coding mode decision method based on machine learning as described in any one of claims 1 to 7 is executed.

Images