基于屏幕内容特性的多叉树划分快速决策算法

Fast Multi-Way Tree Partition Algorithm Based on Screen Content Characteristics

针对H.266中多叉树划分过程计算复杂度高的问题,结合屏幕内容视频不同区域特性设计了一种多叉树划分快速决策算法。首先利用水平、垂直活动度判断屏幕内容空白区域多叉树划分的遍历必要性,其次定义等距子块亮度差变化幅值提前决策多叉树的划分方向,最后利用亮度值平均绝对误差减少多叉树划分的候选模式数量。实验结果表明,与原参考软件VTM10.0相比,所提算法在AI配置和RA配置下都可节省编码时间,Y,U,V三个分量的BD-rate都有所提高。

In response to the high computational complexity of the multi-way tree partitioning process in H.266,a fast decision algorithm for multi-way tree partitioning was designed based on the characteristics of different regions of screen content videos.Firstly,the horizontal and vertical activity levels are used to determine the necessity of traversing a multi-way tree partition in the blank area of screen content.Secondly,the amplitude of brightness difference changes in equidistant sub blocks is defined to determine the partition direction of the multi-way tree in advance.Finally,the average absolute error of brightness values is used to reduce the number of candidate modes for multi-way tree partition.The experimental results show that compared with the original reference software VTM10.0,the proposed algorithm can save encoding time in both AI and RA configurations,and the BD-rate of the Y,U and V components is improved.