期刊
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
卷 26, 期 8, 页码 1502-1515出版社
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TCSVT.2015.2461991
关键词
Coding unit (CU); fast mode decision; High Efficiency Video Coding (HEVC); inter prediction; motion estimation (ME)
资金
- 973 Program [2013CB329004]
- Natural Science Foundation of China [61390510, 61325009]
- Fundamental Research Funds for the Central Universities [WK2100060011]
The emerging High Efficiency Video Coding (HEVC) standard adopts many advanced techniques with flexible combinations, which enables HEVC to achieve about 50% bit-rate reduction for similar perceptual video quality relative to the prior video coding standard H.264/Advanced Video Coding. However, the enormously increased encoding complexity of HEVC inevitably becomes one of the greatest challenges for real-time applications. Among all the factors resulting in the increase in encoding complexity of HEVC, the quad-tree structure for coding units (CUs) with different sizes and accordingly a large number of prediction modes is one critical reason. Thus, it is greatly desired to develop a fast mode decision method for HEVC to reduce the computational complexity. In this paper, considering that HEVC employs the quad-tree structure, and the distortion of each sub-CU can indicate whether the current mode is suitable for current CU, we explore the relationship between the impossible modes and the distribution of the distortions to help the encoder skip checking the unnecessary modes. Besides, since the residual values can reflect the prediction result directly, we propose a method to skip some motion estimation operations according to the distribution of the residuals. Experimental results show that the proposed method can save about 77% of encoding time with only about a 4.1% bit-rate increase compared with HM16.4 anchor, while compared with the fast mode decision method adopted in HM16.4, the proposed algorithm can save about 48% of encoding time with only about a 2.9% bit-rate increase.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据