Noise distance driven fuzzy clustering based on adaptive weighted local information and entropy-like divergence kernel for robust image segmentation

Abstract

Kernel method is an effective way to solve the problem of nonlinear mode analysis, and its key is the selection or construction of kernel function. This paper firstly induced entropy-like divergence by combining Jensen-Shannon/Bregman divergence with convex function, its mercer kernel function called entropy-like divergence kernel is also constructed. Secondly, an adaptive noise distance based on entropy-like divergence kernel and a novel fuzzy weighted local factor of robust fuzzy clustering are presented, and they are also embedded into the objective function of fuzzy C-means clustering with noise cluster. In the end, a novel noise-resistant fuzzy weighed local information clustering based on entropy-like divergence kernel (NEKWFLICM) is proposed, and its convergence is strictly proved by convergence theorem of alternating iteration. Many experimental results delicate that the proposed algorithm has more robust and accurate than a series of existing state-of-the-art Gaussian kernel-based fuzzy clustering-related segmentation algorithms in the presence of high noise.