An un-supervised image segmentation technique based on multi-objective Gravitational search algorithm (MOGSA)

Image segmentation is most important and crucial task in high level image processing. Image segmentation is essentially a multi-objective optimization problem. In this paper an unsupervised image segmentation method is proposed that is based on a nature inspired clustering approach, Multi-objective Gravitational Search algorithm (MOGSA). A variable agent representation is used to encode the cluster centers with different number of clusters. A new fitness function based on multi-objective optimization is proposed to make the search more efficient and faster. Total number of clusters and cluster centers evolves automatically in proposed algorithm. The proposed algorithm considers two objectives, minimize intra-cluster distance, and maximize inter-cluster distance. The proposed method has been applied on standard images and compares the results with K-means clustering, Fuzzy C-Means (FCM) clustering and single objective Gravitational Search Algorithm (GSA). Experimental results showed that Multi-objective Gravitational Search algorithm (MOGSA) perform better than K-means, fuzzy c-means and single objective Gravitational Search Algorithm (GSA).