Contributions au modèle de mélange gaussien pour l'estimation de densités d'histogrammes et applications à l'analyse d'images radar
This thesis is a publication thesis. It presents several of our contributions to the problem of image histogram density estimation.This thesis is divided into three parts. In the first part, we propose two algorithms for estimating the pdf of an image histogram with a mixture model. Both algorithms are designed to estimate the number of modes in a multi-modal histogram and the parameters of each component of the mixture. The algorithms are categorized under two different approaches. The second part of this thesis concerns the generation of test data, and particular emphasis on the generation of artificial histograms. Valid test histograms are essential in the development of mixture pdf estimation algorithms. For this purpose, we first give a formal definition for the concept of overlap between two adjacent components mixtures, and then propose two different algorithms for controlling this overlap. Finally, in the third part of this thesis we present two real applications dealing with target detection and segmentation in SAR (Synthetic Aperture Radar) images. In the first application, we propose an algorithm for the segmentation of small vehicle targets, and in the second we propose an algorithm for ship target detection in SAR images. Six research papers are included in this thesis. Three of them have been published in refereed journals (the Journal of Neural, Parallel and Scientific Computation, the International Journal of Pattern Recognition and Image Analysis and the Canadian Journal of Remote Sensing), two have been submitted to refereed journals (the International Journal of Pattern Recognition and Image Analysis and the International Journal of Remote Sensing), and one is an article published in an international conference (the International Conference on Signal Processing, Application and Technology (ICSPAT'2000)).
- Sciences – Thèses