Abstract: In view of the shortages of conventional texture image segmentation based on Markov random field (MRF) in the wavelet domain, a segmentation method is proposed by combining nonsubsampled Brushlet transform and MRF. Nonsubsampled Brushlet transform is looked on as the feature field of the original image, which makes sure that the high dimensional singularity information of texture image is extracted effectively. And Gauss Markov model is used to compute the arguments of the feature field, which makes sure that the influences of the spectral information and the spatial correlations between pixels on the segmentation result are considered. Experiments show that this algorithm can effectively achieve the texture image segmentation and it is of more great improvement than traditional algorithm in the detection of texture direction information and regional consistency.

Key words: nonsubsampled Brushlet transform; Markov random field (MRF), image segmentation, ICM, MAP criterion