基于模式和时空相关性的运动估计快速算法

H.264标准中的多尺寸块运动估计,在显著提高编码性能的同时,大大增加了其计算量,使得H.264实时编码器的实现面临巨大挑战.本文充分利用视频图像的时空相关性和多尺寸块间的运动相似性,根据运动向量的中心偏置特性,提出了一种运动估计快速算法.该算法通过有效地预测搜索起点,自适应选择搜索模式以及采用二级终止搜索策略等方式,在编码性能相当的情况下,运动估计的速度比全搜索算法提高了95~247倍,比H.264推荐的快速算法提高了4.1~6.3倍.

A Fast Motion Estimation Algorithm Based on Mode and Spatiotemporal Correlation

Compared with the previous standard,the compression performance of H.264 has significantly improved because of different block-size motion estimation.However,encoder using above technology becomes extraordinarily complex.Thus,a real-time video encoder based on H.264 faces enormous challenges.Making full use of the spatiotemporal correlation of video and similar motion between the different-size block,according to the central-biased characteristic of motion vector,a fast motion estimation algorithm appropriated for H.264 is proposed in this paper.By effective prediction of search window center point,adaptive selection of search patterns and termination of the search,the speed of the proposed algorithm speedups 95-247 times than full search and 4.1-6.3 times than the fast motion estimation algorithm recommended in H.264 while maintains similar distortion performance.