Image Segmentation Based On Mean Shift And Improved Ant Clustering Algorithm

As a fundamental and pivotal technology in the digital image processing field, image segmentation can simplify and change the representations of the image effectively. It has been an international hot topic. So far, image segmentation methods can generally be divided into four categories:methods of image segmentation based on edge detection, based on the threshold value, based on the region extraction and the methods which combine the tools of specific theory.To improve the segmentation quality, efficiency and anti-interference ability of color image, a novel approach was proposed in this paper. It incorporates the advantages of the Mean Shift (MS) algorithm and Improved Ant Clustering algorithm. Firstly, it uses Mean Shift to realize preliminary segmentation of the color image, and some regions (about ten) which can preserve the discontinuity characteristics of the image are obtained, then, every region is represented by a node in order to constitute an undirected weighted graph. Like this, image resegmentation problem has been transformed into graph partition problem. An Improved Ant Clustering method is used to solve this problem. Compared with the conventional ant clustering method of image segmentation, it puts forward new similarity calculation way using modified SimRank and improves the maximum number of items that each ant can carry. It can perform globally optimized clustering and shorten the clustering time considerably.To illustrate the applicability and superiority of the method in this paper, many experiments using various kinds of color image have been conducted in this study. Experiment results show that this method can realize color image segmentation effectively, and compared with two relatively new segmentation methods at present, it has improved performance on the image segmentation quality and anti-interference ability. At the same time, because the clustering object of the method in this paper is based on the nodes after the preliminary segmentation using Mean Shift but rather on the pixels of the color image, the computational complexity has been considerably reduced. So, this image segmentation method has good performances and broad application prospect.