Abstract:

The FLICM is an effective algorithm in image segmentation based on the FCM clustering framework. However, it is hard to obtain accurate results for dealing with strong noisedegraded images. By employing the local prior probability of the MRF model, the FLICM algorithm can be improved in two aspects. Firstly, the prior probability is used to weight the dissimilarity function when calculating the fuzzy factor. The refined fuzzy factor takes into account larger scale of neighboring constraints, which makes our algorithm more robust to noise. Secondly, the membership function is further weighted by the prior probability in the process of label determination. Because the neighboring labels must be taken into account in this process, our algorithm can obtain more homogeneous segmentation results. Compared with some other FCMbased algorithms, the proposed algorithm is applied to both synthetic images and real images to demonstrate its strong robustness.

Key words: image segmentation;Fuzzy cmeans clustering;Markov random field model;spatial information