阅读说明：本技术 一种面向vvc的快速帧内预测方法 (VVC-oriented fast intra-frame prediction method ) 是由 魏宏安 林桑 周彬倩 赵铁松 徐艺文 于 2020-07-07 设计创作，主要内容包括：本发明涉及一种面向VVC的快速帧内预测方法,包括以下步骤：步骤S1：构建HAD代价预测模型,预测每种预测模式的HAD代价,按从小到大的排序,并选择前若干个HAD代价的模式初始化CU候选模式列表；步骤S2:通过统计分析候选模式成为最佳模式的概率,优化CU候选模式列表；步骤S3:基于贝叶斯定理,在每个CU深度级别执行CU划分的提前终止,进一步进行VVC的下一步编码流程,从而加快编码时间。本发明实现在保证视频质量的前提下,有效地加快了视频的编码时间。(The invention relates to a VVC-oriented fast intra-frame prediction method, which comprises the following steps: step S1: building an HAD cost prediction model, predicting the HAD cost of each prediction mode, sequencing the HAD costs from small to large, and initializing a CU candidate mode list by selecting a plurality of modes with the HAD costs; step S2, optimizing a CU candidate mode list by statistically analyzing the probability that the candidate mode becomes the best mode; and step S3, based on Bayesian theorem, executing the early termination of CU partition at each CU depth level, and further performing the next encoding flow of VVC, thereby accelerating the encoding time. The invention effectively accelerates the video coding time on the premise of ensuring the video quality.)

1. A VVC-oriented fast intra-frame prediction method is characterized by comprising the following steps:

step S1: building an HAD cost prediction model, predicting the HAD cost of each prediction mode, sequencing the HAD costs from small to large, and initializing a CU candidate mode list by selecting a plurality of modes with the HAD costs;

step S2, optimizing a CU candidate mode list by statistically analyzing the probability that the candidate mode becomes the best mode;

and step S3, based on Bayesian theorem, executing the early termination of CU partition at each CU depth level, and further performing the next encoding flow of VVC, thereby accelerating the encoding time.

2. The VVC-oriented fast intra prediction method of claim 1, wherein step S1 employs a full intra configuration to calculate HAD costs of CUs in different types of video sequences, and divides adjacent reference CUs into UCUs, LCUs and cocus.

3. The VVC-oriented fast intra prediction method according to claim 2, wherein the step S1 specifically includes:

step S11, the HAD cost distribution C (t) of the current CU is predicted by the cost distribution of the adjacent blocks, and the estimation model is as follows:

wherein C is_l(t) and C_u(t) the HAD cost of the LCU and UCU respectively,

step S12: and initializing a candidate list by sorting from small to large and selecting the first N prediction modes, adding a planar mode, a DC mode, a horizontal mode and a vertical mode into a coarse-grained mode selection process, and reducing the prediction modes from 35 to N + 4.

4. The VVC-oriented fast intra-prediction method as claimed in claim 3, wherein the initialization method of step S1 is adopted only in case of simultaneous presence of LCU and UCU in VVC intra-prediction, otherwise the current CU is to be encoded by the original VVC encoder.

5. The VVC-oriented fast intra prediction method according to claim 1, wherein the step S2 specifically includes:

step S21: testing in several different types of sequences, counting the probability that each candidate mode in a candidate list with different CU sizes is selected as the best mode, and proving that accurate prediction can be achieved through the front x mode in the list through a counting result;

step S22: trying a combination of different candidate pattern numbers of a plurality of different CU sizes, wherein the number of candidate patterns of each CU size does not exceed x, comparing the combination with an original candidate pattern scheme of a VVC coder, and selecting a combination with better coding performance and lower computational complexity;

step S23: the CU candidate pattern list is further optimized according to the number of candidate patterns required for different CU sizes listed in the combination obtained in step S22, and the required candidate pattern is selected from the candidate list as the final candidate pattern used in the continuous mode decision process.

6. The VVC-oriented fast intra prediction method according to claim 1, wherein the step S3 specifically includes:

step S31: fitting the RD cost probability density distribution curves of split and undivided CUs with a lognormal function:

wherein r represents RD cost, f_i(r) is the probability density distribution of the RD cost under the condition of i, wherein i has two conditions of s and n, wherein s represents that the CU is continuously split, and n represents that the CU is terminated; μ is the mean, σ²Is the variance;

step S32: calculating error probability P of CU partition_e：

Wherein P (×) is the prior probability of the case of s or n, and P(s) + P (n) -1; th is a threshold value of the RD cost,

step S33: calculating P_eThe value of Th at the minimum, P, where the two curves of the probability density distribution curve of the split CU and the probability density distribution curve of the undivided CU intersect_eMinimum and the intersection value is the optimum threshold Th_opt(ii) a Setting the early termination description of CU partition;

step S34: threshold Th_optShift of (4) to (Th)_optThe rewrite is:

wherein Th_maxIs the maximum threshold for preventing erroneous determination due to an excessively large threshold, and α is an offset value for offline training, and is adapted to different quantization parameters and CU sizes.

7. The VVC-oriented fast intra prediction method as claimed in claim 6, wherein the early termination description of CU partition is as follows:

wherein H_nDenotes the case of terminating CU partition, H_sIndicating that the CU continues to split when the RD cost r is smallEqual to or less than optimum threshold Th_optThe CU partition is terminated.

Technical Field

The invention relates to the technical field of video coding, in particular to a VVC-oriented fast intra-frame prediction method.

Background

VVC is a new generation video coding standard with higher coding efficiency, and many novel techniques have been adopted to improve compression efficiency in the VVC test mode. The concept of deleting the prediction unit and the conversion unit is one of the most important changes, and the coding unit is directly used for predicting and transforming the progress without further division; another key innovation is that the partitioning method adopts a quad-tree with a nested multi-type tree coding block structure instead of the traditional quad-tree structure; in addition, in order to better capture any edge direction, the VVC also expands the number of prediction modes, and adds 32 seed modes on the basis of the original 35 modes.

Disclosure of Invention

In view of the above, an object of the present invention is to provide a VVC-oriented fast intra prediction method, which can effectively accelerate the video encoding time.

In order to achieve the purpose, the invention adopts the following technical scheme:

a VVC-oriented fast intra-frame prediction method comprises the following steps:

step S1: building an HAD cost prediction model, predicting the HAD cost of each prediction mode, sequencing the HAD costs from small to large, and initializing a CU candidate mode list by selecting a plurality of modes with the HAD costs;

step S2, optimizing a CU candidate mode list by statistically analyzing the probability that the candidate mode becomes the best mode;

and step S3, based on Bayesian theorem, executing the early termination of CU partition at each CU depth level, and further performing the next encoding flow of VVC, thereby accelerating the encoding time.

Further, the step S1 adopts a full intra configuration to calculate HAD costs of CUs in different types of video sequences, and divides adjacent reference CUs into UCUs, LCUs and cocus.

Further, the step S1 is specifically:

step S11, the HAD cost distribution C (t) of the current CU is predicted by the cost distribution of the adjacent blocks, and the estimation model is as follows:

wherein C is_l(t) and C_u(t) the HAD cost of the LCU and UCU respectively,

is the corresponding weight, predicted by the weight of the coCU;

step S12: and initializing a candidate list by sorting from small to large and selecting the first N prediction modes, adding a planar mode, a DC mode, a horizontal mode and a vertical mode into a coarse-grained mode selection process, and reducing the prediction modes from 35 to N + 4.

Further, the initialization method of step S1 is only used in the VVC intra prediction when the LCU and the UCU are both present, otherwise the current CU will be encoded by the original VVC encoder.

Further, the step S2 is specifically:

step S21: testing in several different types of sequences, counting the probability that each candidate mode in a candidate list with different CU sizes is selected as the best mode, and proving that accurate prediction can be achieved through the front x mode in the list through a counting result;

step S22: trying a combination of different candidate pattern numbers of a plurality of different CU sizes, wherein the number of candidate patterns of each CU size does not exceed x, comparing the combination with an original candidate pattern scheme of a VVC coder, and selecting a combination with better coding performance and lower computational complexity;

step S23: the CU candidate pattern list is further optimized according to the number of candidate patterns required for different CU sizes listed in the combination obtained in step S22, and the required candidate pattern is selected from the candidate list as the final candidate pattern used in the continuous mode decision process.

Further, the step S3 is specifically:

step S31: fitting the RD cost probability density distribution curves of split and undivided CUs with a lognormal function:

wherein r represents RD cost, f_i(r) is the probability density distribution of the RD cost under the condition of i, wherein i has two conditions of s and n, wherein s represents that the CU is continuously split, and n represents that the CU is terminated; μ is the mean, σ²Is the variance;

step S32: calculating error probability P of CU partition_e：

Wherein P (×) is the prior probability of the case of s or n, and P(s) + P (n) -1; th is a threshold value of the RD cost,

step S33: calculating P_eThe value of Th at the minimum, P, where the two curves of the probability density distribution curve of the split CU and the probability density distribution curve of the undivided CU intersect_eMinimum and the intersection value is the optimum threshold Th_opt(ii) a Setting the early termination description of CU partition;

step S34: threshold Th_optShift of (4) to (Th)_optThe rewrite is:

wherein Th_maxIs the maximum threshold for preventing erroneous determination due to an excessively large threshold, and α is an offset value for offline training, and is adapted to different quantization parameters and CU sizes.

Further, the early termination of CU partitioning is described as:

wherein H_nIndicates termination of CU partitionCase (1), H_sIndicating that the CU continues to split when the RD cost r is less than or equal to the optimal threshold Th_optThe CU partition is terminated.

Compared with the prior art, the invention has the following beneficial effects:

the invention effectively accelerates the video coding time on the premise of ensuring the video quality.

Drawings

FIG. 1 is a flow chart of the method of the present invention;

fig. 2 is a diagram of neighboring reference CUs for intra prediction according to an embodiment of the present invention.

Detailed Description

The invention is further explained below with reference to the drawings and the embodiments.

Referring to fig. 1, the present invention provides a VVC-oriented fast intra prediction method, which includes the following steps:

step S1: building an HAD cost prediction model, predicting the HAD cost of each prediction mode, sequencing the HAD costs from small to large, and initializing a CU candidate mode list by selecting a plurality of modes with the HAD costs;

step S2, optimizing a CU candidate mode list by statistically analyzing the probability that the candidate mode becomes the best mode;

and step S3, based on Bayesian theorem, executing the early termination of CU partition at each CU depth level, and further performing the next encoding flow of VVC, thereby accelerating the encoding time.

In the present embodiment, the HAD cost of CUs in different types of video sequences is calculated using a full intra configuration, and adjacent reference CUs are divided into an upper CU (ucu), a left CU (lcu), and a co-located CU (cocu), as shown in fig. 2.

In this embodiment, the step S1 specifically includes:

step S11, the HAD cost distribution C (t) of the current CU is predicted by the cost distribution of the adjacent blocks, and the estimation model is as follows:

wherein C is_l(t) and C_u(t) the HAD cost of the LCU and UCU respectively,is the corresponding weight, predicted by the weight of the coCU;

step S12: the candidate list is initialized by sorting from small to large and selecting the first 10 prediction modes, and a coarse-grained mode selection process is entered by adding a planar mode, a DC mode, a horizontal mode and a vertical mode together, so that the prediction modes are reduced from 35 to 14, and the coding complexity is reduced.

Preferably, in VVC intra prediction, some blocks may not have neighboring blocks, which results in that the parameters required by the model cannot be fully obtained, so the above candidate mode initialization method is only used if the LCU and the UCU are present at the same time, otherwise the current CU will be encoded by the original VVC encoder.

In this embodiment, the step S2 specifically includes:

step S21: testing in several different types of sequences, counting the probability that each candidate mode in a candidate list with different CU sizes is selected as the best mode, and proving that accurate prediction can be achieved through the first 2 modes in the list through a counting result;

step S22: trying a combination of different candidate mode numbers of a plurality of different CU sizes, wherein the number of candidate modes of each CU size is 1-2, comparing the combination with an original candidate mode scheme of a VVC coder, and selecting a combination with better coding performance and lower computational complexity;

step S23: the CU candidate pattern list is further optimized according to the number of candidate patterns required for different CU sizes listed in the combination obtained in step S22, and the required candidate pattern is selected from the candidate list as the final candidate pattern used in the continuous mode decision process.

In this embodiment, the step S3 specifically includes:

step S31: fitting the RD cost probability density distribution curves of split and undivided CUs with a lognormal function:

wherein r represents RD cost, f_i(r) is the probability density distribution of the RD cost under the condition of i, wherein i has two conditions of s and n, wherein s represents that the CU is continuously split, and n represents that the CU is terminated; μ is the mean, σ²Is a variance, and can be calculated by maximum likelihood estimation.

Step S32: calculating error probability P of CU partition_e：

Wherein P (×) is the prior probability of the case of s or n, and P(s) + P (n) -1; th is a threshold value of RD cost for determining whether a CU needs to be split.

Step S33: calculating P_eThe value of Th at the minimum, P, where the two curves of the probability density distribution curve of the split CU and the probability density distribution curve of the undivided CU intersect_eMinimum and the intersection value is the optimum threshold Th_opt(ii) a Setting the early termination description of CU partition; the early termination of CU partitioning can therefore be described as:

wherein H_nDenotes the case of terminating CU partition, H_sIndicating that the CU continues to split when the RD cost r is less than or equal to the optimal threshold Th_optThe CU partition is terminated.

Step S34: threshold Th_optShift of (4) to (Th)_optThe rewrite is:

wherein Th_maxIs the maximum threshold for preventing erroneous determination due to an excessively large threshold, and α is an offset value for offline training, and is adapted to different quantization parameters and CU sizes.

The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made in accordance with the claims of the present invention should be covered by the present invention.