摘要
根据图像序列运动矢量的空间分布特性,针对H.264视频压缩提出了一种基于十字交叉和局部全搜索相结合的整像素运动估计快速算法。该算法首先在一维方向上对一个宏块中的所有子块同时进行运动搜索,只在最后的步骤选中的4个点的周围进行5×5的局部全搜索,减少了搜索点数,降低了运算复杂度。同时由于该算法将各种模式的搜索集成在一个搜索过程,并且搜索中前后两点在空间上是相邻的,非常有利于VLSI实现。利用JM12.0平台对该算法进行了仿真。结果表明,对于CIF序列,该算法较全搜索算法在保证图像质量的前提下,节省了92%的搜索时间。
By analyzing the spatial distribution character of motion vectors in video, a new fast integer-pixel motion estimation algorithm is proposed based on cross search and local full search. Firstly, the algorithm performs motion estimation in 1-D direction for all the subblocks in a macroblock, and then make 5 × 5 full search around 4 points at last. The procedure reduces the searching points and the algorithm integrates searching procedures for all modes into one process, and searches in a way that the neighbors are adjacent in spatial location, so it is very appropriate for VLSI implementation. Experiments based on JM12.0 show that when pro- cessing CIF frame sequences, the efficient motion estimation algorithm can save 920% searching time compared with the full search algorithm with negligible quality loss.
出处
《数据采集与处理》
CSCD
北大核心
2009年第3期309-312,共4页
Journal of Data Acquisition and Processing
基金
国家自然科学基金(60425413)资助项目
关键词
运动估计
H.264
超大规模集成电路
motion estimation
H. 264
very-large-scale integration (VLSI)circuit