A New Method for Image Segmentation based on MultiObjective Differential Evolution Fuzzy Clustering

Image segmentation is one of the most important and difficult steps in machine vision problems and achieving the desired results often requires satisfaction of different objectives. One approach to face this situation uses multi-objective fuzzy clustering of pixels in the feature space. This paper proposes a new strategy for search within the family of multiobjective differential evolution algorithms with the purpose of finding optimal partitions of pixels. Based on this¡ all of the encoded clusters in the population of one generation are clustered and each centroid of one donor vector is made with centers from a unique cluster. Through the search space dimension is reduced and searching for each centroid focused on the different area of input space while cluster centers of one fuzzy partition preserve separation. The performance of the proposed method is compared with two other multi-objective fuzzy clustering methods to segment a number of images from the Berkeley segmentation database. Visual and quantitative evaluations show that the proposed method has a better match with the ground truths than the other methods.