基于运动矢量空间相关性的H.264分像素运动估计

H.264 fractional pixel motion estimation based on spatial correlation of motion vector

随着整像素运动估计快速算法的发展,分像素运动估计的计算量在运动估计中所占比重越发明显。为了减少分像素运动估计的计算量,提出了一种利用运动矢量空间相关性来预测整像素运动块,对整像素运动块进行分像素搜索过程跳过的分像素运动估计方法。实验结果表明,该算法与全分像素搜索算法结合使用,在基本保持搜索精度不变的情况下,比单纯的全分像素搜索算法减少60%左右的分像素搜索点。该算法可与其他快速分像素搜索算法结合使用,以获得更好的编码性能。

With the development of fast integer pixel motion estimation algorithms, the proportion of the computational cost of fractional pixel motion estimation in that of the whole motion estimation has become more and more significant. To reduce the complexity of fractional pixel motion estimation, a novel method of H.264 fractional pixel motion estimation based on the spatial correlation of motion vector is proposed. It predicts Integer Pixel Motion Block（IPMB） using the spatial correlation of motion vector. It skips the fractional pixel search process for IPMB. Experimental results show that combined with the full fractional pixel search algorithm, the proposed algorithm can reduce 60% of fractional pixel search points while maintaining the search accu- racy basically compared with the pure full fractional pixel search algorithm. The proposed algorithm can be combined with other fractional pixel search algorithm to get better coding performance.