Fuzzy C-Means Clustering With Neighbor Information Constraint Using Sparse Self-Representation

Fuzzy c-means (FCM) clustering is a significant yet efficient unsupervised learning methods in many fields such as image segmentation, pattern recognition, etc. However, the traditional FCM algorithm cannot perform well at segmentation on images with vague boundaries especially in the presence of noises. To address this problem, by means of the sparse self-representation technique and the incorporation of the neighbor information, a novel FCM clustering method is put forward, which is called fuzzy c-means clustering with neighbor information constraint using sparse self-representation (SSRFCM_N). The main idea of the proposed SSRFCM_N is two fold: 1) besides the traditional cluster center regarding the global information of similarity, another center with respect to the local information is introduced into the objective by using the sparse self-representation technique; and 2) to consider the data distribution adequately, the neighbor information constraint is also incorporated into the objective, contributing to a better accuracy as well as the good robustness to the noise. Finally, experiments on different images show that SSRFCM_N is effective and more competitive than state-of-the-art clustering methods.