A random field approach to unsupervised texture image segmentation

An unsupervised random field approach, which involves local and long range information in determining the class of target image blocks, in texture segmentation problem is introduced in this work. Like Markov random field (MRFs) approaches, the proposed method treats each of the image blocks as a site and attempts to assign an optimal class label to each of it. Unlike MRF,s, which involve only local information extracted from a small neighborhood, in addition to the local neighbors, our method allows a few long range blocks to be involved in the labeling process, in an attempt to alleviate the problem of assigning different class labels to disjoint regions of the same texture and the problem of over-segmentation due to the lack of long range interaction among the neighbors and the distant blocks. The proposed method requires no a priori knowledge of the number and types of region.