Nouvelle méthode en cascade pour la classification hiérarchique multi-Temporelle ou multi-capteur d'images satellitaires haute résolution

This paper describes a method dedicated to multi-resolution, multi-date and eventually multi-sensor classification based on explicit statistical modeling through hierarchical Markov random field modeling based on quad-Tree. The proposed approach consists in a supervised Bayesian classifier that combines a joint class-conditional statistical model for pixelwise information and a hierarchical Markov random field for spatio-Temporal and multiresolution contextual information fusion based on the Marginal Posterior Mode (MPM). The aim is to recursively maximize the posterior marginal at each pixel, which associates the most probable class label given the entire input information. Within this framework, an interesting novel element of the proposed approach is the use of multiple quadtrees in cascade, each associated with a new image in the available set in order to characterize the correlations associated with distinct images in the data set.