优化预测运动矢量的快速运动估计算法  

Novel fast motion estimation algorithm based on optimizing predictive motion vector

作　　者：闫敬文[1] 余见[1] 屈小波[1] 张晓玲[1] 

机构地区：[1]厦门大学通信工程系

出　　处：《光学精密工程》2007年第10期1622-1627,共6页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.60472081);航空科学基金资助项目(No.05F07001)

摘　　要：提出了一种优化预测运动矢量的快速运动估计算法。在对预测运动矢量研究的基础上,根据序列图像中运动矢量的中心-中值偏置分布特性和矢量间的时空相关性,结合运动矢量的相似度分析,选用中心、中值和时间相关的三个矢量作为基本预测矢量。设置相似门限来减少由三个空间相邻块预测矢量带来的大量冗余信息,对算法中关键的门限技术进行了改进。实验结果证明,本文算法对各种类型的运动序列都有很强的自适应性,在保持搜索准确度的同时,可大幅度提高运动估计的速度,其平均搜索速度是FS的208倍,明显优于PMVFAST的146倍、MVFAST的77倍、DS的55倍,提高了视频压缩中现有的运动估计算法的性能。A new Optimizing Predictive Motion Vector Fast Motion Estimation Algorithm （OPMVFMEA） is proposed. Based on the center-median-biased distributed characteristic of the motion vector of real-world sequences and the high space-time correlation of adjacent block＇s motion vector, combined with similar analysis of the motion vectors, the center, median, forward vector are selected as the basic predictive motion vector to predict the current one, then the similar threshold is set to reduce the redundant information from the three space adjacent block motion vectors. In addition, the adaptive threshold to enable half-stop is also represented. Experiment results show that the algorithm is able to adapt to all types of video sequences and can offer a high performance of PSNR. The search speed of the algorithm is faster than that of the existing well-known algorithms. For case examined, it is about 208 times faster than that of FS in average, which is superior to that of 146 times for PMVFAST, 77 times for MVFAST and 55 times for DS. Therefore, the algorithm improves the performance of existing motion estimation algorithms.

关 键 词：运动估计 块匹配 视频压缩 预测运动矢量 冗余信息 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]