摘要： 通过分析序列编码结果，根据已编码帧预测模式分布和序列的运动信息，对运动特性不同的宏块选择不同的运动搜索方案。结合提前退出策略以及分层的搜索模式，提出一种新的运动搜索算法。实验结果表明，算法编码质量接近全搜索算法，算法执行时间小于全搜索算法，在速度和质量上都优于钻石搜索算法、新3步法、六边形搜索算法，适用于不同运动特性的视频序列。

关键词: 视频编码, 运动估计, 宏块

Abstract: This paper proposes a novel motion search algorithm. This algorithm sorts the video and search pattern by motion feature respectively, and codes the video with corresponding search pattern adaptively. In order to accelerate the coding speed, it accepts hierarchical search strategy and the early termination strategy in the algorithm. Experimental results show that the execute time of the algorithm is less than full search algorithm with similar visual quality and bit rate. The method can be applied to various video sequences with different motion feature and is superior to conditional search algorithm, such as NTSS, DS and HEXS in both encoding speed and encoding quality.

Key words: video coding, motion estimation, macroblock

中图分类号: 

• TN91.21