I. Introduction
The original concept of the active contour model or snakes [1] is proposed by Kass. The author proposes an energy function and the specifically defined energy function will only be minimized when the active contour stops on the edge of the object. This kind of model which contains the gray level gradient information in the energy function is called edge-based model. Later, Mumford and Shah [2] propose a region-based energy function which replaces the edge term with two terms about an approximation function. But the problem is the minimization of the energy function is too difficult. In order to simplify the Mumford-Shah energy function, different constraints on the approximation function are made, for example the piecewise linear constraint [3], piecewise smooth constraint [4][5] or piecewise constant constraint [6]. Among the above models, the most widely-used one is the piecewise constant model proposed by Chan and Vese. They solve the Mumford-Shah minimization problem by level set method and propose a segmentation model which can segment the object without clear edge [6]. Thereafter, many variations of the Chan-Vese model have been proposed. Some researches focus on the improvement of the Chan-Vese model [7][8], some researches focus on adding the shape model into the segmentation process [9][10][11][12] and some researches focus on simultaneously segmenting multi objects using multiphase level set [13][14][15]. There are also some models that hybridize the edge-based method and region-based method to improve the segmentation results on the specific images [16].