基于矢量预测的快速运动估计搜索算法

基于H.264视频编码标准的编解码过程中,运动估计的时间大概要占总编码时间的70%(1个参考帧)到90%(5个参考帧)。对于H.264标准的新特点,传统的全搜索算法的精度高,但计算量太大,不能应用于实时处理;经典的菱形等算法搜索模式简单,易于实现,但容易陷入局部无穷小。采用了一种基于运动矢量预测的快速运动估计搜索算法。该方法首先利用运动矢量的时、空间相关性得到预测矢量,然后利用非对称十字型搜索确定运动估计的起始点,最后采用经典的菱形算法进行运动估计。实验结果表明,相比UMHexagonS快速搜索算法,该算法能够在码率增加不超过1%,信噪比下降不超过0.1 dB的情况下,运动估计速度有较大提高。

A fast motion estimation search algorithm based on Vector prediction

Motion Estimation(ME)can consume 70%(1 reference frame) to 90%(5 reference frames) of encoding time of the H.264.Traditional full-search algorithm of inter coding has high precision,but the computation cost is so large that it can not be used for real-time processing.Classical Diamond Search(DS) algorithm is simple,but is likely to be trapped into a local minimum.We used a quick search algorithm based on motion vector prediction.This algorithm utilizes the temporal and spacial correlation of the motion vector for obtaining the predictive vector,then uses unsymmetrical-cross search to determinate a starting point for motion estimation;and uses diamond search algorithm at last to end up the motion estimation.The experimentation result show that:compared with UMHexagonS,the algorithm can increase motion estimating speed greatly with an increase of coding rate no more than 1% bite-rate and descending of SNR no more 0.1 dB.