Robust entropy-based symmetric regularized picture fuzzy clustering for image segmentation

Abstract Symmetric regularized picture fuzzy clustering is a new fuzzy clustering method, and it is difficult to choose its weighting fuzzy factor m and lacks certain robustness to noises or outliers. To this end, this paper proposes a robust entropy-based symmetric regularized picture fuzzy clustering with spatial information constraints for noisy image segmentation. The idea of maximum entropy fuzzy clustering is firstly introduced into symmetric regularized picture fuzzy clustering, and we can obtain an entropy-based symmetric regularized picture fuzzy clustering method to avoid selecting weighted fuzzy factors m. Meanwhile, it has clear physical meaning. Then, spatial neighborhood information of current clustering pixel is embedded into its clustering objective function, and a new robust fuzzy clustering algorithm for image segmentation is obtained to enhance the ability to suppress noise. In the end, the convergence of new robust clustering algorithm is proved by Zangwill theorem. Many experimental results show that the proposed algorithm can achieve higher segmentation performance than existing picture fuzzy clustering.