Stability-integrated Fuzzy C means segmentation for spatial incorporated automation of number of clusters

Abstract

Fuzzy C Means clustering, one of the predominant segmentation algorithms, requires prior knowledge of number of clusters in the image and is sensitive to noise and outliers. Determining the number of clusters and including spatial information to basic Fuzzy C Means clustering are done in numerous ways. Literature reveals that either number of clusters is defined or spatial information is incorporated. In the proposed work, spatial information and cluster determination are integrated based on the concept of stability. Implementation of split and merge algorithm to find the number of clusters is done based on the modified Sylvester’s theorem in the context of positive definite functions. Experiments are performed on synthetic and real images and the number of clusters determined is validated using validation indices. Results show that correct clusters are classified with robustness to noise.