A robust local data and membership information based FCM algorithm for noisy image segmentation

This paper presents a technique for incorporating local data and membership information into the standard fuzzy C-means (FCM) algorithm. The objective function associated with the technique consists of a modified version of the standard FCM function plus a weighted regularized FCM-like one. In the first function, the Euclidian pixel-to-cluster distances are computed using the original data. However, in the second one, they are computed by replacing the original data by locally smoothed one to reduce additive noise. Both distances are also modified to account for the distances in the pixel neighborhood. In both functions, to incorporate the local membership information, the resultant pixel-to-cluster distance is weighted by the reciprocal of the average of the membership to this cluster in the pixel vicinity. Results clustering synthetic and medical images are presented. The performance of the proposed robust local data and membership information FCM (RFCM) is compared with the standard FCM, local spatial information based FCM (SFCM), and data and local data and membership weighted FCM (LDMWFCM).