Fuzzy C-means algorithm incorporating local data and membership information for noisy medical image segmentation

This paper presents an approach to incorporating both local data and membership information into the standard fuzzy C-means (FCM) clustering algorithm. In this approach, the standard FCM function is regularized by a weighted fuzzy c-means term. However, in this term, the pixel-data is replaced by local pixel-data average for the computation of the distance from the cluster center. Also, both the distances of the standard FCM and the additional regularizing term are weighted by the reciprocal of local membership average. Therefore, clustering a pixel is influenced by the pixel-data and both data and membership information of its immediate neighboring pixels. This leads to enhance the performance in clustering noisy images and to bias the clustered images toward piecewise homogenous regions. Simulation results are presented to compare the proposed algorithm with the standard FCM and several local data and membership based FCM algorithms.