An MRF approach to unsupervised change detection

An approach to the automatic analysis of the difference image for change detection in multitemporal remote sensing images is proposed. This approach is based on a technique that exploits the expectation-maximization (EM) algorithm for the estimation of the density functions associated with both the changed and unchanged pixels in the difference image. Then, on the basis of such estimates, an automatic method for the unsupervised analysis of the difference image is described. The method makes use of Markov random fields (MRFs) for modeling the spatial-contextual information included in the neighborhood of each pixel. Experimental results confirm the effectiveness of the proposed approach