Image Segmentation by Improved Conical Area Evolutionary Algorithm

Abstract

AbstractDecomposition based multi-objective evolutionary algorithms are suitable to solve the multi-objective optimization problems and becoming popular. For image segmentation, clustering is an unsupervised learning method. Most clustering algorithms results are highly dependent on the initial cluster center and it is important to optimize the segmentation process. In this paper, we used an improved conical area evolutionary algorithm named CAEA-II for the optimization process to search the optimal cluster centers. We consider two objectives namely, minimize the compactness of intra-cluster and maximize the separation of inter-cluster to define the initial optimal cluster centers. Xie Beni index (XBI) calculates the separation and compactness of cluster centers and Average Inter-Cluster Separation (AIS) calculates the clusters minimum overlapping. CAEA-II uses the cone decomposition method to construct the Pareto frontier through optimization of multiple conical sub-problems simultaneously. Then we used Davies-Bouldin Index (DBI) to define the optimal solutions for cluster centers. Experiment results proved that the proposed technique for image segmentation gives better results than the Possibilistic Clustering Algorithm (PCA) and single-objective optimization (SOO).KeywordsCAEAPossibilistic clustering algorithmBi-objective optimizationImage segmentation