一种可伸缩的预测性快速运动向量搜索算法  

Controllable Predictive Search Algorithm for Fast Block Motion Estimation

作　　者：杨天武[1] 彭强[1] 诸昌钤[1] 

机构地区：[1]西南交通大学计算机与通信工程学院,四川成都610031

出　　处：《西南交通大学学报》2005年第1期30-34,共5页Journal of Southwest Jiaotong University

基　　金：铁道部科技研究开发项目(2003X040-A)

摘　　要：基于钻石搜索的特点,提出了搜索距的概念.通过对不同搜索距的检测点特性的研究,并吸收了PMVFAST算法中"见好就收"的思想和初始预测候选运动向量集的概念,提出了一种新的可伸缩的预测性十字方块快速运动搜索算法.该算法以检测点的块失真特性统计为基础,制定出精确可控的可适应搜索准则.大量模拟实验表明:该算法能获得更好的视频质量.同时,拥有良好的搜索速度伸缩性,在允许图象质量有0.05dB降低时,搜索速度能提高1.5～2.5倍.Based on the characteristics of the diamond search, the concept of search length was brought forth. A novel controllable predictive cross-square fast search algorithm was proposed through investigating the characters of checking points with different search lengths. The proposed algorithm takes concepts of stop when good enough spirit and MIPC (multiple initial predictor candidates) in PMVFAST (predictive motion vector field adaptive search technique) algorithm. In the proposed algorithm, controllable predictive search criteria are set down based on the statistical characteristics of block distortion of checking points. Simulated experiments show that the proposed algorithm can obtain a better video quality than the other search algorithms. In addition, it has a good control capability of search speed, and its search speed can increase 1.5 to 2.5 times with less than 0.05 dB degradation of video quality.

关 键 词：视频 运动估算 十字搜索 钻石搜索 H.263 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]