An Improved MRF Based Unsupervised Change Detection Method

Traditional unsupervised change detection algorithms based on simple MRF model assume that subimages applied to extracting features are homogeneous, but that is not always true and causes low accuracy. Based on the fields correlation Markov random field (CMRF) model, an adaptive algorithm is proposed in this paper. The labeling is obtained through solving a MAP (Maximum a posterior) problem by ICM (Iteration Condition Model). Features of each pixel are exacted by using only the pixels currently labeled as the same pattern. With the adapted features, the new labeling is obtained. The satisfied experimental confirm the effectiveness of proposed techniques. Although the proposed method has been presented in the specific context of the analysis of multitemporal remote-sensing images, it could be used in any change detection application requiring the technique based on the difference image