An Improved Approach to Image Segmentation Based on Mumford-Shah Model

In this paper, based on M-S model and level-set method, a new method for initializing level-set function and hierarchical constant segmentation is proposed in order to overcome the shortcomings in the Chan-Vese model. First, by improving the initial level set function, the process of re-initializing level set function in traditional method is eliminated, and the initial conditions are easier to handle. Secondly, this paper presents a hierarchical constant segmentation method for multi-phase segmentations, using estimated energy to determine whether the sub-regions need further segmentations. By evolving only one level set curve at each segmentation stage, our algorithm speeds up the segmentation significantly. Finally, various numerical experiments on both artificial and real images are done to validate the proposed methods