面向H.264的快速运动估计算法

Fast motion estimation algorithm for H.264

最新的H.264视频编码标准的性能比先前的相关标准有很大的提高。但增加了编码器的复杂度。尤其是使用穷举搜索算法的时候,运动估计和模式判定占用相当多的编码时间,从而降低了编码器的速度。该文提出了一种面向H.264的快速整像素运动估计算法—UCMRGS算法。该算法使用多分辨率网格搜索策略,结合运动矢量预测和非对称交叉搜索,有效地降低了运动估计的计算量,同时保证非常好的视频质量。实验结果表明,该文提出的快速运动估计算法,可以大大提高MPEG-4AVC/H.264编码器的运行速度,相比穷举运动搜索算法视频质量没有下降。

The new H.264 video-coding standard has a significant performance benefit versus former standards. Unfortunately, the advanced coding features including variable block-size motion compensation and multiple reference frames incur a considerable increase in encoder complexity especially when using the straightforward full search algorithm, mainly in regards to motion estimation and mode decision. This paper describes a new motion estimation algorithm for fast H.264 coding based on the generalized motion vector predictors, the hybrid unsymmetrical-cross multi-resolution grid search (UCMRGS). The algorithm takes advantage of the correlations among variable blocks to considerably reduce the computational cost at the encoder while at the same time giving similar, and in some cases better, visual quality compared with the brute force full search algorithm. The algorithm mainly relies upon very robust and reliable predictive techniques with parameters adapted to the local characteristics combined with UCMRGS.