A novel hybrid model for two-phase image segmentation: GSA based Chan-Vese algorithm

Abstract The active contours without edges model of Chan and Vese (Chan and Vese, 2001), which has been accepted for two-phase image segmentation is one of the most widely-used methods. It is a region-based segmentation model that utilizes the techniques of curve evolvement and the level set method. Chan–Vese model is a strong and flexible method that is able to segment many types of images compared to other active contours. Nevertheless, improper initial contours may reveal the problem of the Chan–Vese model getting stuck in a local minimum. This situation often provides poor results for the Chan–Vese model. Particularly, this problem occurs in the images that have large intensity differences between local and global structures. In this paper, we present a novel hybrid approach to the Chan–Vese algorithm to bring a solution to the problem of segmentation of these images. The proposed approach is based on the Gravitational Search Algorithm (GSA) developed in Rashedi et al. (2009). The idea is to arrange the fitting energy minimization problem according to a heuristic optimization technique and provide satisfactory segmentation outcomes regardless of the choice of the initial contour. The proposed model has been tested on both several images taken from Weizmann dataset and suitable medical images for the local minima problem. Experiments on the suitable test images prove that the proposed GSA based Chan–Vese model is more accomplished and more robust when compared to the conventional Chan–Vese algorithm. The test results also denote that the proposed algorithm requires much smaller number of iterations (%75 less) to converge than the conventional Chan–Vese algorithm.