Classification et sélection de caractéristiques de textures Utilisation d'algorithmes automatiques supervisés de sélection d'attributs pour la classification d'images

Image's experts use different kind of attributes to represent texture information. We propose a methodology to automatically choose the best texture models using a feature selection algorithm. Therefore we compare the efficiency of several recent algorithms. The algorithms evaluation is performed using classification error rates and heuristics. We demonstrate the interest of such a methodology on Brodatz and satellite images.