Low Complexity Quadtree based All Zero Block Detection Algorithm for HEVC
High Efficiency Video Coding (HEVC) is a proposed video compression standard, a successor to H.264/MPEG-4 AVC (Advanced Video Coding). It aims to substantially improve coding efficiency compared to H.264/AVC High Profile, in which transform unit (TU) has been added as one of the key advanced features, at the expense of traversal computation of transform and quantization quadtree. To release the computation burden brought by TU, all-zero block (AZB) detection algorithms can be introduced, which early terminate transform and quantization with almost no PSNR loss. However, the previous works can’t support the tree structure based (depth-2 or more) new standard. And by setting the thresholds with sum of absolute difference (SAD) only, they can not work directly in case of intra prediction or other criterion based inter prediction. This paper proposes a quadtree based variable sized all-zero block detection scheme for high efficiency video coding which mainly consists of all-zero block detection for prediction unit (PU) and quadtree based upbottom zero node detection. Firstly, 3232 with depth-3 integer transform and quantization based all-zero block detection has been derived by gaussian distribution modeling, of which the approach can also be extended to sum of absolute transformed differences (SATD) and sum of squared differences (SSD). Further more, the chroma based condition has also been proposed by study of the color component correlation in this paper. Moreover, the up-bottom quadtree based adaptive to variable size and depth all-zero node detection has been designed. By implementing in the HEVC reference test model (HM), evaluation results illustrate the proposed algorithm can save total encoding time 4％-25％ averagely with at most -0.24dB BDPSNR and +3.55％ BDBR loss.
Jia Su Yiqing Huang Lei Sun Shinichi Sakaida Takeshi Ikenaga
Waseda University, Fukuoka, Japan Science & Technology Research Laboratories, NHK, Tokyo
国际会议
2011亚太信号与信息处理协会年度峰会(APSIPAASC 2011)
西安
英文
1-10
2011-10-18（万方平台首次上网日期，不代表论文的发表时间）