Influence de l'estimation des paramètres de texture pour la classification de données complexes

Anthony Fiche (1) , Jean-Christophe Cexus (1) , Arnaud Martin (2) , Ali Khenchaf (1)

1 E3I2 - Extraction et Exploitation de l'Information en Environnements Incertains
2 CORDIAL - Human-machine spoken dialogue

This paper shows a classification of data based on the theory of belief functions. The complexity of this problem can be seen as two ways. Firstly, data can be imprecise and/or uncertain. Then, it is difficult to choose the right model to represent data. Gaussian model is often used but is limited when data are complex. This model is a particular case of α-stable distributions. Classification is divided into two steps. Learning step allows to modelize data by a mixture of α-stable distributions and Gaussian distributions. Test step allows to classify data with the theory of belief functions and compare the two models. The classification is realized firstly on generated data and then on real data type sonar images.

Mots clés: alpha stable; belief functions; sonar image

Domaines: Intelligence artificielle [cs.AI]; Traitement du signal et de l'image [eess.SP]; Traitement du signal et de l'image [eess.SP]

Anthony Fiche, Jean-Christophe Cexus, Arnaud Martin, Ali Khenchaf. Influence de l'estimation des paramètres de texture pour la classification de données complexes. Extraction et Gestion des Connaissances, Jan 2011, Brest, France. pp.10h. ⟨hal-00657521⟩