基于方向自适应菱形搜索的运动估计算法  

Directional adaptive diamond search for motion estimation

作　　者：李荣[1] 张其善[1] 杨东凯[1] 

机构地区：[1]北京航空航天大学电子信息工程学院,北京100191

出　　处：《北京航空航天大学学报》2008年第9期1065-1069,共5页Journal of Beijing University of Aeronautics and Astronautics

基　　金：国家自然科学基金资助项目(60602046)

摘　　要：基于对运动矢量分布特性的研究,提出了一种基于方向自适应菱形搜索的运动估计算法.该算法对搜索起始点进行预测;设置"双阈值",针对匹配块提前中止搜索;并根据运动特征,自适应地选择小菱形模板和4种新型的方向自适应菱形模板,具有强烈的搜索方向性.实验结果证明,该算法不仅大幅度地减少了平均搜索点数,而且在一定程度上提高了重建图像的信噪比,其搜索速度和精度均优于传统的快速运动估计算法.Based on the research of motion vector distribution, a novel directional adaptive diamond search （DADS） algorithm was proposed. This algorithm adopted three strategies： the initial search point prediction strategy, the early search termination strategy, and the directional adaptive diamond search strategy. The initial search point prediction strategy made the predicted motion vectors close to true values and avoided them falling into the local optimal. The early search termination strategy detected the matching macroblock according to two fixed thresholds, and terminated the current search immediately. The directional adaptive diamond search strategy adopted the small diamond search pattern （SDSP） and four novel directional adaptive diamond search patterns （ DADSP）, and the search patterns were utilized adaptively based on the type of motion activity. Experimental results show that the proposed algorithm not only significantly reduces the average search points, but also improves the search accuracy to a certain extent, and it excels other traditional fast block matching algorithms in speed and veracity.

关 键 词：运动估计 块匹配 视频 方向自适应菱形搜索 

分 类 号：TN919.81[电子电信—通信与信息系统]