【作者】 李振亚； 宋建斌； 李波；

【Author】 LI Zhenya+, SONG Jianbin, LI Bo Digital Media Laboratory, Department of Computer Science, Beihang University, Beijing 100083, China

【机构】 北京航空航天大学计算机学院数字媒体室；

【摘要】 基于块的运动估计是视频压缩国际标准中广泛采用的关键技术,运功估计快速算法也一直是本领域研究的热点,本文通过分析运动估计中运动矢量搜索陷入局部最优的原因,提出了一种分区域的分级搜索策略,有效的降低了搜索陷入局部最优的可能性,在此基础上结合有效的搜索起点预测、提前终止准则等策略,设计出一种新的运动估计快速算法,该算法在几乎不降低编码器编码质量的情况下,能有效的减小编码器的运算量,提高编码速度。更多还原

【Abstract】 Motion estimation is the most important part in exploiting the high temporal redundancy between successive frames and is also the most time consuming part in the hybrid coding framework, which is specifically serious in H.264 because of multi prediction modes, multi reference frames, and higher motion vector resolution. As a result, algorithms on fast motion estimation are always hot research spot. Many fast integer block-matching algorithms have been focused on the search strategies with different steps and search patterns in order to reduce the computation complexity and maintain the video quality at the same time. But in some cases, search process is likely to drop into a local minimum in the early stages. To solve this "local-minimum" problem, some previous works predicted the motion vector from the neighboring matching blocks or coordinate block and then searched from the predicted motion vector. Another method is that global search is performed with more search points that can cover the overall search area. And search points were selected from sparser grids in the overall search window in order to reduce the number of search points in the first search step. In this paper we analyse the reason why the search points get into a local minimum,and finds a effective solution to it,which is called hierarchical search strategy in multi search areas.This solution devide the whole search window into several parts and perform a coarse to fine search process in every part.Based on this search stragety,we proposed a a new algorithm on fast motion estimation which also combines initial search point prediction and early termination technical.The experiment results show that this fast search algorithm could save about20%encoding time compared with UMHexagonS algorithm,with very small loss in the video quality.更多还原