一种面向快速全局运动估计的渐进精细阈值方法

Progressive Fine Granular Threshold Method Oriented to Fast Global Motion Estimation

快速全局运动估计的关键在于全局运动区域与局部运动区域的分割,其难点在于阈值的设定。提出一种面向快速全局运动的渐进精细阈值方法,该方法分为两步:第一步,用一个将统计特性与均值阈值相结合的亮度残差阈值模型来大致划分局部运动区域与全局运动区域,从而得到全局运动估计区域的近似集合;第二步,使用一种运动矢量残差分级阈值技术在能量残差函数最小化迭代过程中逐步细化全局运动像素点集合,最后分离出完整的全局运动区域,从而实现快速的运动估计。

The method of the fast global motion estimatin focuses on separating global motion region and local motion region.The key point of this method is how to set the appropriate threshold.This paper proposed a progressive fine granular threshold technique to solve this problem.The technique is divided into following two steps：firstly,distinguishing coarsely the global motion region and local motion region by using a luminance error threshold model integrating statistics method and average threshold method;secondly,during the iteration for minimizing error,refining the set of the global motion pixels by using a hierarchical threshold method based on motion vector error.By this means the whole global motion region can be obtained and the fast global motion estimation can be achieved.