基于最优位置计算的快速亚像素运动估计

A Fast Sub-Pixel Motion Estimation Algorithm Based on Best Position Calculation

亚像素运动估计是提高视频压缩性能的有效方法,但当前普遍采用的亚像素全搜索算法不仅运算复杂度极高而且随亚像素精度的增加插值存储开销成指数级增长.该文提出了一种快速算法,该算法首先利用整像素运动估计的中间结果直接计算出最优亚像素位置,并在此基础上将最终搜索点数减少到2个以下,避免了传统方法的逐点搜索比较,不仅大幅度提高了搜索速度,且插值存储开销趋近于零.实验结果表明该算法以极小的搜索代价取得了与全搜索相当的效果.

Sub-pixel motion estimation is efficient to improve performance of video compression. Yet the full sub-pixel search algorithm adopted widely nowadays not only seems time-consuming but also leads to exponential growth memory spending for interpolation image as the increasing of sub-pixel accuracy. A fast algorithm is proposed in this paper, which calculates the best position in sub-pixel level directly instead of traditional checking points one by one, then checks two points at most based on the calculated position. As a result, search speed is enhanced greatly without additional memory cost. Experiments show that the algorithm obtains similar PSNR to full sub-pixel search at very low computation cost.