Coarse-fine Surrogate Model Driven Preference-based Multi-objective Evolutionary Fuzzy Clustering Algorithm for Color Image Segmentation

When applied to image segmentation, most existing multi-objective evolutionary clustering algorithms usually consider the information at the pixel level. Sometimes ignoring the region information of the image may lead to unideal segmentation performance. Moreover, these multi-objective evolutionary clustering algorithms require expensive computation for evaluating fitness functions. In order to improve the segmentation performance and time efficiency for image segmentation, a coarse-fine surrogate model driven preference-based multi-objective evolutionary fuzzy clustering (CFS-PMEFC) algorithm is proposed in this paper. First of all, the coarse-fine surrogate model and the preference information derived from the user are introduced to design multi-objective evolutionary clustering framework, which can improve the evolutionary efficiency and quality of clustering centers. Then, Mahalanobis distance-based fitness functions fusing the region information of the image are constructed to further promote the segmentation performance. In addition, a clustering index with the region information of the image is designed to select the optimal solution from the final solution set of CFS-PMEFC. Experiments on Berkeley images show that CFS-PMEFC greatly improves the accuracy of image segmentation and meanwhile reduces the computational cost.