Unsupervised texture segmentation using 2-D AR modeling and a stochastic version of the EM procedure

The problem of textured image segmentation upon an unsupervised scheme is addressed. Until recently, there has been few interest in segmenting images involving possible complex random texture patterns. It is also a fact that most unsupervised segmentation techniques generally suffer from the lack of information about the correct number of texture classes. Therefore, this number is often assumed known a priori. On the basis of the so-called SEM (Stochastic Expectation Maximisation) algorithm, we try to perform a reliable segmentation without such prior information, starting from an upper bound for the number of texture classes. The image model first assumes an autoregressive (AR) structure for the class-conditional random field, and in a further step, a Markovian structure of the region process. The application of this method on a textured mosaic is presented.