运动分割需要估计出每个运动的运动模型参数和运动支持区。为准确地确定运动支持区，在最大后验边缘概率(MPM MAP)算法的基础上，提出了一种新的基于区域收缩的运动分割算法，用于确定运动支持区。该算法先以属于某种运动概率最大的像素为备选像素，然后通过区域收缩选定备选像素密度最大的区域为支持区范围。此外，还提出了一种新的运动模型参数初值的估计方法，并将差分图像包围盒的确定和区域收缩相结合，用于运动模型参数的初值估计，该方法先估计运动区域，再估计运动模型参数,并通过运动分解、合并和消亡来获得准确的运动个数。实验结果表明，该方法是有效的。

Motion segmentation needs to estimate the model parameters of every motion as well as its supporting region. On the basis of maximum posterior marginal probability (MPM MAP) algorithm this paper presents a new algorithm based on region shrinking to locate the supporting area. In this algorithm pixels of maximum probabilities belonging to a motion are considered to be candidate pixels for supporting region. Then the region shrinking algorithm is used to determine the region of maximum density of the candidate pixels to be the range of supporting area. Moreover, this paper presents a new approach combining the bounding box’s defining with region shrinking to estimate the initial parameters of motions. By motion dividing, motion incorporation and motion elimination the accurate number of motions can be obtained. The results of experiments show the validity of this method.