基于起点预测的不连续十字形快速搜索算法

Discrete cross fast search algorithm based on initial search point prediction

提出了一种基于起点预测的不连续十字形块搜索算法。该算法首先利用图像帧的时间相关性和空间相关性预测出一个点作为不连续十字形搜索模板的初始中心点,再根据运动矢量概率分布的中心十字偏置特性,采用不连续十字形搜索模型,对运动矢量进行估计,得出最佳运动估计结果。在保证图像质量的前提下,该搜索算法与完全搜索算法(FS)、三步搜索算法(TSS)和钻石搜索算法(DS)相比,搜索速度显著提高。

This paper analyzed the spatial and temporal correlation of motion trend of the current block, studied the motion vector distribution, and proposed a novel adaptive discrete cross-search algorithm based on initial search point prediction. It utilized a novel initial pixel point selection strategy to improve the accuracy while keeping the computation cost less than other conventional algorithm. Experiments show that the algorithm is able to fit for all types of video sequences adaptively in spite of the degree of the motions,and it is better than the traditional fast motion estimation algorithms such as FS, TSS, and DS, in terms of speed and quality.