Multiscale Image Segmentation Using Markov Random Field and Spatial Fuzzy Clustering in Wavelet Domain

In this paper, a multiscale image segmentation algorithm based on Markov random field and spatial context fuzzy clustering in wavelet domain is presented. At the determination of pixel label stage, the feature field of image is described by Gaussian mixture model, the label field of image is characterized by Markov random field, according to the Bayesian criterion, the initial label of wavelet coefficients from coarse to fine scale is determined. At the image segmentation stage, the modified fuzzy c-means objective function with locally spatial constraint is introduced by the initial label of different scale wavelet coefficients. The algorithm is put forward to overcome the shortcomings of standard fuzzy clustering, which is extremely sensitive to noise and lacks of spatial constraints. The performance of the proposed method is compared with that of the spatial domain Markov random field model and the conventional fuzzy clustering segmentation algorithm. Experiments on simulated images have demonstrated the efficiency of the proposed approach, such as accurately locating image edges, correctly identifying different regions and immunizing to noise.