Improved Template-Matching-Based Fruc Method In Inter-Frame Video Coding

S. C. Song, H. W. Guo,C. Zhu,Y. B. Gao,Y. B. Lin,J. H. Zheng

2018 13TH IEEE INTERNATIONAL SYMPOSIUM ON BROADBAND MULTIMEDIA SYSTEMS AND BROADCASTING (BMSB)(2018)

引用 1|浏览10
暂无评分
摘要
As a new inter prediction mode, Frame Rate Up-Conversion (FRUC) is introduced for exploring the future video coding. In FRUC, for a coding unit, the motion information can be derived by the Template Matching (TM) technique. However, in the TM-based FRUC, the prediction direction, including forward, backward, or bi-directional prediction, is determined only based on the initial motion information without further using pixel information of the current block. This may decrease the efficiency of the FRUC mode. To solve this problem, in this paper, three selection schemes on prediction direction are proposed based on template matching distortion and rate distortion cost, respectively. In the first scheme, the relationship in terms of template matching distortion is obtained among different prediction directions. Accordingly, based on this relationship, the prediction direction can be determined both at encoder and decoder, and no additional bit needs to be transmitted to the decoder. In the second scheme, the prediction direction is determined based on rate-distortion optimization, and two additional bits are transmitted to the decoder for each block. Finally, by combining the above two schemes, the third scheme is proposed to further enhance coding efficiency while reducing coding complexity. Our proposed algorithm is integrated into the Joint Exploration Model 5.0.1 platform. Experimental results show that average Bjontegaard delta rate (BD-rate) savings of 0.23% (Y), 0.47% (Y), 0.51% (Y) for three schemes are achieved by using the proposed three schemes, respectively, with no or a little increase of computational cost.
更多
查看译文
关键词
Inter-frame video coding, Template matching, FRUC mode, Prediction direction selection
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要