A CMRF-Based Approach to Unsupervised Change Detection in Multitemporal Remote-Sensing Images

Simple MRF model based method usually suffers from less inaccuracy because it assumes that each subimage used to estimate features is homogeneous. In this paper, an adaptive algorithm based on the fields correlation Markov random field (CMRF) model is proposed. The labeling is obtained through solving a MAP problem by ICM. Features of each pixel are calculated by using only the pixels currently labeled as the same pattern, while the new labeling is obtained by using the adapted feature. The satisfying experimental results in change detection of multitemporal remote-sensing differencing images confirm the effectiveness of proposed techniques.