Multi-objective Evolutionary Clustering with Complementary Spatial Information for Image Segmentation

When existing multi-objective evolutionary clustering algorithms is applied to image segmentation, it can not obtain satisfactory segmentation performance on an image corrupted by noise due to no consideration of any spatial information derived from the image. Based on the complementarity of the local spatial information and the non local spatial information of the image, these two kinds of spatial information are introduced into a cluster validity function, and a novel objective function with complementary spatial information is constructed, and then a multi-objective evolutionary clustering algorithm with complementary spatial information for image segmentation is proposed. In order to reduce human intervention, the variable string length real coded technique is adopted to determine automatically the number of clusters during the evolving process. Natural image segmentation experiments show that the proposed method not only can obtain satisfactory segmentation performance on noisy images, but also can be suitable for many types of noisy images.