Shadowed C-Means for Image Segmentation Using Local and Non-local Spatial Information

This paper introduces some new image segmentation methods in the framework of shadowed c-means clustering. By implanting the local and non-local spatial information in the membership value estimation procedure, we propose the Local Spatial Shadowed C-Means (LSSCM) algorithm, Non-local Spatial Shadowed C-Means (NLSSCM) algorithm and their combination - L+NLSSCM. Compared to traditional fuzzy c-means and shadowed c-means based approaches, the proposed image segmentation algorithms can obtain better segmentation results on test images. It is observed the proposed algorithms can effectively tackle the overlapping among segments and the noise problem in images.