Hierarchical Bayesian models for multi-sensor processing
Dissertation defended on October 19th, 2007, at ENSEEIHT Toulouse (France) in candidacy for the degree of Doctor of Philosophy of the National Polytechnic Institute of Toulouse
In order to process the mass of information collected in many applications, it is necessary to propose new processing methods to exploit the "multi-sensor" feature of the observed data. The subject of this thesis consists in studying algorithms of estimation in a multi-sensor context where several signals or images resulting from the same application are available. This problem is of great interest since it makes it possible to improve the estimation performances compared to an analysis that would be carried out on each signal independently of the others. In this context, we have developed methods of hierarchical Bayesian inference to perform segmentation of multiple signals and to analyze hyperspectral images. The use of Markov Chain Monte Carlo methods allows one to overcome the difficulties related to the computational complexity of these inference methods.
