A KERNEL METRIC AND SPATIAL INFORMATION DRIVEN SIMILARITY MEASURE FOR FUZZY C-MEANS CLUSTERING

Abstract

In this paper we present a new idea of fuzzy c-means based image segmentation algorithm with a novel similarity measure. Unlikely the Euclidean metric based conventional similarity measure here proposed method hired kernel induced distance measure together with spatial information influenced Euclidean norm by exploiting the significant of both, and computation of this proposed similarity measure is totally free from any adjusting parameter that overcomes trade off ambiguity. Simultaneously, the proposed method could preserve image details as well as suppress image noises via the use of local label information by the addition of the modification of similarity function which is determined by the relationship of the neighborhood pixels. Experiment has been carried out on both synthetic and different kind of real images, and result of proposed algorithm show marked segmentation performance, resistance to noisy images with complex distribution both quantitatively and qualitatively.