Sélection de modèles pour la classification supervisée avec des SVM (Séparateurs à Vaste Marge). Application en traitement et analyse d'images.

摘要

The objective of this thesis is to define learning systems based on SVM with good performance. These systems should take into account that the problems related to image processing and analysis may enter into conflict with the operational difficulties of SVM. Many of these issues are part of the broader framework of data mining, the definition of decision-making in real-time optimization of difficult problems and combination of sets of decision functions. The approaches proposed in this thesis to solve problems of various kinds can be used in other areas where the same problems are encountered.