Bayesian approach to automatic change detection

In this paper, an approach to the selection of the decision threshold for change-detection techniques based on the difference image is proposed. This approach, unlike classical ones, allows the decision threshold to be selected in an entirely automatic way. In particular, an iterative technique is proposed, which exploits the expectations-maximization (EM) algorithm for the estimation of the statistical terms associated with the gray levels of changed and unchanged pixels in the difference image. Then, on the basis of such estimates, two different strategies for the selection of the decision threshold are presented: one is based on the Bayes rule for minimum error (BRME); the other is based on the Bayes rule for minimum cost (BRMC).