基于边缘光流法向分量的运动区域划分

为克服传统运动区域分割算法对光流求解正确度的依赖,提出了一种基于光流法向分量的运动区域分割算法.该算法利用有区别的同区域判决函数对各个边缘点及其邻域进行光流法向分量的相关性聚类.利用Matlab软件对合成图像和真实图像进行了仿真.结果表明,在参数选择合理的情况下,该算法能够正确地对图像序列中的运动区域进行分割;并且由于仅对于边缘点处的光流法向分量进行处理,算法速度也比基于完全光流矢量的算法有所提高.

Motion Area Division Method Based on Normal Component of Edge Optical Flow

To lessen the accuracy demand of traditional motion area division methods, a motion segmentation algorithm based on normal component of optical flow is proposed. Edge-discriminating same area decision function is used to give out the relevance measurement between each edge point and its surrounding area, which is the base for further clustering. Simulation on both synthesized and real image sequences was made. It is proved that under appropriate selection of parameters, the novel algorithm can give correct division result. Moreover, compared to those methods using complete optical flow vector, the new method costs less time because that only normal component of the edge point' optical flow is evolved in the algorithm.