用于运动位移估值的新匹配准则

New matching criterion for motion displacement estimation

在视频编码中，位移估值是编码速度的瓶颈。除了研究快速搜索算法外，匹配准则也是研究的热点。为了提高运动位移矢量的精度，减少估值用的运算量，利用人眼对物体边缘运动较敏感的特性，提出了一种新匹配准则——边缘像素点分类准则。在Ｓｏｂｅｌ边缘检测的基础上，对处于物体边缘位置的像素点根据其亮度值进行分类，一类为匹配点，另一类为不匹配点。将搜索区域内含边缘匹配点最多的子块判为匹配块。计算机模拟表明，利用边缘像素点分类准则进行位移估值得到的重建图像，质量比目前常用的平均绝对误差准则有显著改善，同时大大地减少了运算量。

Because displacement estimation is the bottleneck of video coding, matching criterion becomes a research focus as well as the studing of fast algorithm. In order to get accurate result of displacement esimation with low computation cost, according to the fact that human eyes are highly sensitive to the motion of edge, this paper proposes a new matching criterion called the edge pixels classification criterion. After edge detection, the pixels in the edge are classified into the following categories: matching pixels and mismatching pixels. The best match within a search area is the block with the highest number of matching pixels. The simulation results show that this new criterion performs significantly better than the widely used mean absolute difference (MAD) criterion, while the computation complexity is much lower.