A probabilistic generative model for unsupervised invariant change detection in remote sensing images

In this paper we present a probabilistic generative model for the change detection problem. Generative models represent homogeneously all relevant variables in a specific domain by a joint probability distribution. The proposed model explicitly represents the image formation process (including possible brightness transforms between images or registration errors) and is invariant to affine changes in pixel intensities or small georegistration errors. There are several benefits from such theoretical formulation: all the modeling assumptions are explicit and the method to solve the change detection problem is not intrinsic to the formulation. The use of probabilistic models also leads to sound and well-known statistical techniques for problems like parameter estimation or regularization. The experimental results confirm the validity of the approach.