方向性多层抽样继承排除全搜索算法

运动估计是视频压缩中的关键技术，是视频编码中的主要开销。全搜索算法是最简
单可靠的块匹配算法。本文在传统全搜索算法的基础上，提出一种方向性多层抽样继承排除全搜索算法(DMSSEA)。本文算法通过数理统计的方法设置判别条件，在全搜索中引入提前终止；并通过抽样计算降低计算量，提高计算灵活度；通过分层判别提高判别效率；通过对图像方向性的利用，提高了终止效率。该算法在保证运动估计全局最优的同时大大减少了搜索点数，提供高清晰度的视频编码。
关键词运动估计方向性 多层次 抽样 继承排除
Directed multi-level sample successive elimination Algorithm LI tao， LIU ying jie， LI yangxi， MA yide (School of information science and engineering， Lanzhou University， Lanzhou， 730000) Abstract Motion estimation (ME) plays an important role in video coding system. It is the main calculation. In this paper ， a new algorithm， directed multi-level sample successive elimination， is proposed for fast optimal block matching in motion. By using set judgment ， in an aim to reject a checking candidate as early as possible; by using sample reduce the computation cost， improve the finesse; by using the direction judgment ， improve the efficiency of early stop; thus reducing computational load as much as possible. This algorithm decreased the number of matching evaluations， and ensure the estimation is the best of the whole.
Keywords motion estimation ; direction; multi-level; sample; successive elimination.