Unsupervised segmentation of textured images

In this paper, a new unsupervised segmentation algorithm for textured images is developed. The proposed algorithm utilizes the characteristic of the Markov random fields (MRF) for modeling the contextual information embedded in image formation. Textured images are regarded as realizations of the stationary Gaussian MRF on a two-dimensional square lattice, and are modeled by the conditional autoregressive (CAR) equations with a second-order noncausal neighborhood. A hypothesis test over two radially masked areas is involved in our algorithm to detect the location of boundaries. The hypothesis assumes that these two areas belong to the same class of textures, and CAR model parameters are estimated in a minimum-mean-square-error (MMSE) sense. The hypothesis is tested by the analysis of variance (ANOVA)-like technique, and if it is rejected, a measure of dissimilarity is accumulated on the rejected area. This approach produces potential edge maps. From these maps, the boundaries among various textures can be detected without microedges. The performance of the proposed algorithm is demonstrated by some experiments using real textures as well as synthetic ones. The experiments show that the proposed algorithm can produce satisfactory segmentation without any a priori information.