Fuzzy model-based clustering and its application in image segmentation

This paper proposes a novel fuzzy model-based unsupervised learning algorithm with boundary correction for image segmentation. We propose a fuzzy Generalized Gaussian Density (GGD) segmentation model and the GGD-based agglomerative fuzzy algorithm for grouping image pixels. The merits of algorithm are that it is not sensitive to initial parameters and that the number of groups can be estimated via the validation technique. To minimize the objective function of the model, we define a dissimilarity measure based on the Kullback-Leibler divergence of the GGDs that computes the discrepancy between GGDs in the space of generalized probability distributions. To effectively segment images with various textures, we propose a two-stage fuzzy GGD segmentation algorithm. The first stage adopts the proposed fuzzy algorithm to obtain initial segmentation and the second stage improves initial segmentation by image boundary correction. Experimental results show that our proposed method has a promising performance compared with existing approaches.