论文部分内容阅读
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。
In video coding, the estimation of the displacement is the bottleneck of the coding speed. In addition to researching fast search algorithms, matching criteria are also hot topics. In order to improve the precision of motion displacement vector and reduce the computational complexity, a new matching criterion, edge pixel classification criterion, is proposed based on the characteristics of human eye which are sensitive to the edge movement of objects. On the basis of Sobel edge detection, the pixels at the edge of the object are classified according to their brightness values, one is the matching point and the other is the mismatch point. The sub-blocks with the most edge matching points in the search area are judged as matching blocks. Computer simulation shows that the reconstructed image obtained by using the edge pixel classification criterion for displacement estimation has a significantly improved quality compared with the commonly used average absolute error criterion and greatly reduced the computational complexity.