Research On Fast Coding Unit Partition Algorithm In H.266/VVC

As a new-generation coding standard to replace High Efficiency Video Coding,Versatile Video Coding achieves better coding efficiency at the cost of a large increase in computational complexity,and its coding time is nearly 10 times longer than that of HEVC.The newly introduced multi-type tree(MTT)division structure in VVC enables the Coding Unit(CU)to be flexibly split to satisfy various texture patterns.But this structure causes a significant increase in computational load of WC.Reducing the computational complexity while controlling the loss of coding quality is the goal of rapid optimization of VVC.According to the-complex process of obtaining the optimal partitioning result of coding blocks in VVC intra prediction,this paper designs a multi-type tree fast division algorithm and an early termination algorithm for coding block division by using the texture characteristics of sub-blocks and the depth of adjacent blocks.The main contents of this article are as follows:1.This paper studies the relationship between the selection of multi-type tree partitioning and the texture characteristics of CU sub-blocks.The selection of binary tree partitioning in the horizontal or vertical direction is closely related to the difference of sub-block directional gradients.The paper designs a multi-type tree fast partitioning algorithm by using the texture features of sub-blocks.First,we use the difference of sub-block directional gradients to skip redundant binary tree partitioning,and defines SBGD(gradient difference of sub-block)to measure the gradient difference of CU sub-blocks.By comparing horizontal SBGD and vertical SBGD value to skip unnecessary binary tree partitioning to reduce computational complexity.In addition,based on the correlation between the ternary tree partitioning and the difference of the standard deviation of the sub-blocks,we use the difference of the standard deviation of the sub-blocks to skip the redundant ternary tree partitioning.For binary tree partitioning and ternary tree partitioning in multi-type tree partitioning,this paper uses different texture features of sub-blocks to skip redundant partitioning,the proposed algorithm can save 41%coding time on average,and the BDBR is only increased by 0.55%,BD-PSNR is-0.04.2.Within a frame of image,the texture features of adjacent CUs are very similar,and the depth difference is also small.The paper analyzes the correlation of the depth of the adjacent CUs after the completed encoding,and proposes an early termination decision algorithm for CU partitioning based on adjacent block depths.Considering that in a frame of image,except for the leftmost and uppermost CUs,the remaining CUs have adjacent left and upper CUs directly adjacent to them.The method uses the depth of the coded CUs on the left side and the upper side of the current CU to refer to the depth range of the current CU,and determines whether it can terminate the current CU division in advance.In intra-frame prediction,this paper uses early termination of the current CU partition to avoid calculating the redundant ratedistortion cost at larger depths,the proposed algorithm can save 27.80%coding time on average,and the BDBR is only increased by 1.07%,BD-PSNR is-0.05.