Multivariate Markovian gamma distributions for multitemporal sequences of SAR images

This paper introduces a family of multivariate gamma distributions characterized by a Markovian correlation structure. These distributions are interesting for detecting changes between multi-temporal sequences of synthetic aperture radar images. The parameters of these distributions can be estimated by using the maximum likelihood principle. This estimation procedure allows one to obtain a change indicator between each pair of images. A performance comparison with maximum likelihood estimators derived from a bivariate gamma distribution model is conducted. The gain of performance provided by the Markovian multivariate model with respect to a bivariate model is emphasized on synthetic and real images.