摘要： 提出了一种应用在H.264中的快速运动估计算法(TAMV)。该算法根据运动矢量自身特性，将宏块分为3种运动类型：大运动快，中等运动块和小运动块。对每类宏块自适应地选择一种或者几种编码模式进行预测，有效地减少了由7种宏块(亚宏块)模式所引起的计算量；同时选用精确而且快速的方向性菱形模板，对不同的模式采用不同的搜索策略。实验分析表明，该算法搜索精度接近于全搜索算法，而搜索时间只为全搜索算法的5.4%～10.5%。

关键词: H.264, 运动矢量特征分类, 方向性菱形模板

Abstract: This paper proposes a novel motion estimation algorithm based on texture analysis of motion vector in H.264 (TAMV). According to motion vectors distribution and correlation within modes, TAMV proposes an efficient inter mode decision approach to reduce the number of candidate modes of macro block. And it selects a directional diamond search pattern in the motion estimation while maintaining the coding efficiency. Experiments show that TAMV can obtain good PSNR performance and very fast speed as well as low bit rates.

Key words: H.264, Texture analysis for motion vector, Directional diamond search pattern