H.264/AVC中基于搜索区域划分及评估的运动估计

Motion Estimation Based on the Partition and Evaluation of the Search Area in H.264/AVC

目前视频编码器中的快速运动估计搜索算法(如三步法、钻石搜索法)搜索点数少、速度快,但易陷入局部最优。另外一些混合模板搜索算法(如UMHexagons算法)较为有效地克服了这一缺点,然而这种算法需要处理较多的搜索点数导致搜索速度较慢。为了达到在减少搜索陷入局部最优可能的同时尽量降低搜索点数,提高搜索速度的目的,提出了PEOSA算法。首先利用预测运动矢量来区分出需要重点搜索的区域和较为次要的区域,其次依据重要性的不同对区域采用不同的搜索方案。对比实验的结果表明,本文算法在基本不影响图像质量的情况下,使运动估计时间减少了43.84%,提高了搜索速度。在视频实时编解码方面具有较高的实际应用价值。

Although the search speed of fast motion estimation algorithms such as Three Step Search and Diamond Search is fast, it is easy to trap into local-minimum which caused the false motion vector prediteor. UMHexagons solved the problem well; however, this algorithm is still needed to search too many points and therefore slowed down the search speed. In order to obtain a good balance between search speed and quality performance, this paper proposed PEOSA algorithm. It divided the search area into four part firstly, then evaluated the probability of the best motion vector in every part, and finally searched every part through different search scheme. The experiment results showed that the PEOSA algorithm save more than 43.84% search time compared with UMHexagons ; while the averaging PSNR loss was less than 0. 017dB for all tested sequences . The new search algorithm holds high value in real time coding and decoding.