Kernel non-local shadowed c-means for image segmentation

In order to apply successfully the fuzzy clustering algorithms like shadowed C-means (SCM) to image segmentation problems, the spatial information related with each pixel in the image should be carefully calculated and appended to the clustering algorithms. In this paper, the non-local spatial information calculation is introduced to SCM. Because the data in the kernel space demonstrate more linearly-separable shape and the distances calculated in it shows the property of robust to noise and outliers, the proposed clustering algorithm is conducted in the kernel space (aka feature space) mapped from the original space by some implicit mapping functions defined in the kernel functions. Simulations results on some noise images and the comparison with traditional methods demonstrate the efficiency and superiority of the proposed new approach.