Abstract: Among many modified fuzzy c-mean (FCM) algorithms， the adaptive spatial fuzzy c-means (ASFCM) clustering algorithm is quite advantageous， as it has the adaptive parameters and changes the structure of spatial penalty to make the objective function continuous， but it cannot restrain the large noise contained by infant brain MR images. In response to this issue， we improve the ASFCM algorithm with non-local weights and the kernel function， which is named as the improved ASFCM algorithm with kernel function and non-local weights. Then， the FCM algorithm， RFCM algorithm， ASFCM algorithm and the algorithm we proposed are used to segment the clinical infant brain MR images with different kinds and intensities of noise. Results show that the segmentation accuracy and denoising ability of the proposed algorithm are greatly improved compared with the other three algorithms， and our algorithm has obvious advantages for the infant brain MR image segmentation.

Key words: image segmentation, kernel function, ASFCM, infant brain, MR image