Fuzzy Clustering And The Application On Image Segmentation

Image segmentation is the technology and process that an image is divided into a number of non-overlapping regions which have consistency in properties and extract the interesting target. Image segmentation is one of the key technologies in digital image processing and also is a classical problem in computer vision. Image segmentation is the foundation of image analysis and image understanding. Image segmentation has been successfully used in many fields such as computer vision, pattern recognition, target tracking and medical images processing etc.Because of the effect of various factors in the imaging procedure, similarity and uncertainty exist between background and objectives to be extracted. Fuzzy theory and fuzzy image processing technology is especially suitable to handle these issues with uncertainty. Fuzzy clustering method is an important theoretical branch of image segmentation technology. The fuzzy c-means (FCM) is a classic fuzzy clustering method which is widely used in many fields. The clustering came down to a nonlinear programming problem with constraints, the final fuzzy partition of dataset is obtained by optimizing the objective function.Based on the basic theory of FCM, we focus on the disadvantages of the algorithm and the problems found in image segmentation. Our research priority is how to improve the initial clustering center and the robustness of FCM. Two solutions are proposed in this paper. Research of the dissertation is concentrated on the following aspects:(1) A novel semi-supervised fuzzy c-means clustering method is proposed in this paper. A constraint set is composed of a small amount of labeled data. We use the center of constraint set as the initial clustering center of FCM to obtain the initial membership matrix. This modification can improve convergence rate, reduce iteration numbers and avoid the algorithm trapped in local optimum.(2) When FCM is used to solve image segmentation problems, effect of spatial distribution information of neighborhood pixels to the performance of robustness is considered in this paper. A part of neighbor pixels is used as spatial constraint term added in the objective function. Solving the modified function by Lagrange multiplier, the final iterate optimize formula is obtained. Experiments on standard images and real image segmentation proved that the modified method can improve the robustness of the algorithm with no significantly increase in computing time.