一种基于子块纹理差异的VVC快速CU划分算法

A fast CU partition algorithm for VVC based on sub-block texture difference

针对多功能视频编码(versatile video coding, VVC)由于在四叉树(quadtree, QT)划分的基础上引入多叉树(multi-type tree, MTT)划分结构导致编码复杂度过高的问题,本文提出了一种基于编码单元(coding unit, CU)子块纹理特征差异的VVC快速CU划分算法。该算法能够有效降低VVC的编码复杂度,缩短编码时间。首先通过不同划分方向的子块的纹理复杂度差异提前判断MTT划分的方向,跳过不必要的MTT划分方向;然后根据当前CU的相邻子块间的纹理差异判断是否跳过三叉树(ternary tree, TT)划分,以进一步减少候选列表中划分模式的数量。实验结果表明,在全帧内(all intra, AI)的编码配置下,与官方测试平台VTM-7.0相比,本文算法平均能够节省47.24%的编码时间,BDBR(Bjontegaard delta bit rate)仅增加1.26%。

Aiming at the problem of high coding complexity due to the introduction of multi-type tree(MTT) partition structure based on quadtree(QT) partition in versatile video coding(VVC),a fast coding unit(CU) partition algorithm based on the difference of texture features of CU sub-blocks is proposed in this paper.This algorithm can effectively reduce the coding complexity of VVC and shorten the coding time.Firstly, through the text complexity difference of sub-blocks in different partition directions, the direction of MTT partition is judged in advance, and unnecessary MTT partition directions are skipped;Then, according to the texture difference between adjacent sub-blocks of current CU,it is further determined whether to skip the ternary tree(TT) partition, so as to further reduce the number of partition modes in the candidate list.The experimental results show that under the coding configuration of all intra(AI),compared with the official test platform VTM-7.0, the proposed algorithm can save 47.24% of the coding time on average, and the Bj?ntegaard delta bit rate(BDBR) only increases by 1.26%.