高性能视频编码帧间预测的单元划分优化算法

An Effective CU Splitting Algorithm in Inter Prediction of HEVC

新一代的高性能视频编码(HEVC)通过应用各种先进技术来大幅提高视频编码的性能,然而,这些方法也大大增加了整个编码过程,尤其是预测编码阶段的计算复杂度。该文提出一种应用于高性能视频编码帧间预测的快速单元划分算法。该算法基于HEVC灵活的四叉树结构,根据被编码单元的运动特性来确定所编码信息的阈值,并以此作为单元划分的提前终止条件,从而减少了不必要的单元划分操作和率失真代价的相关计算,达到节省编码时间和降低编码复杂度的目的。实验结果表明,在峰值信噪比(PSNR)损失仅为0.0418 d B的情况下,所提算法可以平均降低46.1%的编码时间。

High Efficiency Video Coding（HEVC） provides better compression performance by adopting many new techniques. However these tools also increase the computational complexity of the prediction, which consumes most of encoder computations. This paper proposes an effective splitting algorithm in inter prediction of HEVC. Based on the motion homogeneity of coding unit, a certain threshold is used to decide whether the coding unit should be split into smaller ones. So the unnecessary calculation decreases rapidly. By skipping some specific sub-CUs, the coding complexity is dramatically improved. Experimental results show that the proposed technique can save 46.1% coding time in average with negligible loss of coding efficiency, and the decrease of PSNR is only 0.0418 d B.