Fuzzy clustering techniques, especially Fuzzy C-Means clustering method (FCM), is a popular algorithm widely used in the images segmentation. However, as the conventional FCM doesn't optimize data in feature space and doesn't involve any spatial information, it is sensitive to the noise. In the paper, we presented a novel FCM clustering algorithm based on kernel spatial information to segment the images. The kernel-induced distance is used as a substitute of the traditional Euclidean distance then the objective function includes the spatial penalty term, which makeups the impact of the neighborhood pixels on the center pixel. At the same time, the parameter can be automatically learned with the regulatory factor. The proposed algorithm is utilized to synthetic and simulation MR images and it is more robust to noise and outline than the other FCM methods.