Shape prior based foreground segmentation with local rotation and structural changes

Abstract

Shape prior based segmentation method has the advantage of robustly tracking moving object. But, it can not deal with local rotations or changes of object structure. In this paper, we propose an algorithm that can accurately segment the moving foreground object with local rotation and object structure change from the low resolution gray scale images based on Schoenemann's method. Two new terms, turning angle and a distance function, are added to the energy function. When estimating the similarity between the prior shape template and the object contour, turning angle is used to measure the curve match, which is rotational invariance and translational invariance. To deal with the object structure change, a distance function is added to the energy function. The experimental results show that the improved new energy function can robustly segment and track the foreground object with local rotations and structure change.