【摘要】 运动估计中菱形算法是一种高效的搜索方法,但它没有利用视频序列中运动矢量的中心偏置分布特性和相邻宏块在时间和空间上的相关性,因此还有很大的空间可以进一步提高该算法的性能。在充分利用运动矢量的中心偏置分布特性和相邻宏块运动矢量的相关性的基础上,提出了一种基于自适应预测的菱形搜索算法(APDS),与传统的起点预测方法不同的是,该算法对所有宏块搜索起点的预测并不是按照一种固定的方法求取,而是通过分析其相邻宏块运动矢量的特性来自适应求得搜索起点和决定搜索策略。实验证明,同传统的菱形法相比,这种方法提高了搜索准确度,同时大幅降低了搜索时间,是一种高效率的搜索算法。更多还原

【Abstract】 The diamond search is one kind of effective searching algorithm,but it doesn’t utilize the center-biased distribution characteristic of the motion vectors and the spatial correlation of the adjacent macro-blocks,thus there exists large space to improve the performance of this algorithm.In this paper,a diamond search algorithm based on adaptive prediction(APDS) is proposed.It takes full advantage of the center-biased distribution characteristic and the spatial correlation of the adjacent macro-blocks.Compared to traditional prediction methods,APDS algorithm employs adaptive prediction method to obtain an initial search point and one of searching schemes for each macro-block,other than to get a fixed initial search point and the same searching scheme.Simulation results show that,compared to diamond search(DS),APDS not only improves the precision,but also reduces the complexity.In the mean time,APDS obtains good image quality with almost equivalent PSNR to full search algorithm(FS).更多还原