Multivariate image segmentation with cluster size insensitive Fuzzy C-means

This paper describes a technique to overcome the sensitivity of fuzzy C-means clustering for unequal cluster sizes in multivariate images. As FCM tends to balance the number of points in each cluster, cluster centres of smaller clusters are drawn to larger adjacent clusters. In order to overcome this, a modified version of FCM, called Conditional FCM, is used to balance the different sized clusters. During the clustering process, the ratios between the cluster sizes are determined and a corresponding condition is calculated. This condition value balances the influence of objects from larger clusters to smaller clusters. Experiments with the cluster size insensitive FCM (csiFCM) on different numerical datasets, synthetic and real multivariate images for different number of clusters and cluster sizes show the improvement compared to FCM and FMLE.