In order to more effectively segment noise-corrupted brain MRI images

a kernelized clustering algorithm using fuzzy Markov random field (MRF) model is proposed. The proposed algorithm is implemented by incorporating the MRF based spatial constraints as a regularization term to the objective function of the kernelized fuzzy C-means clustering（FCM）. The spatial connectivity modeled by the Gibbs distribution is actually formulated as a normalized Gaussian radius basis function (GRBF)

and very similar to the kernel function used to measure the intensity feature of image data. Due to the introduction of fuzzy information in the spatial constraints

the MRF and GRBF based clustering algorithm improves the segmentation model and usually outperforms the conventional intensity based FCM method and the corresponding kernelized clustering method. The modified algorithm can incline the solution to a piecewise smoother segmentation result. Experiments on synthetic data

simulated and real clinical MR images and the result comparisons with ground truth show the proposed algorithm is superior to its rivals and is effective to segment MR data corrupted by noise.