Combinaison hiérarchique de systèmes d'inférence floue : application à la reconnaissance en-ligne de chiffres manuscrits

Abstract : Recently, an increasing gain of attention has been paid on classifiers combination to treat complex problems such as handwriting recognition. These kind of classifiers, based on complementarity principles, increase the performances of a recognition system by limiting errors of a single classifier. In this article, we describe a hybrid system based on the hierarchy of two different king of modeling. A first level model intrinsically the classes by fuzzy prototypes and deduces a pre-classification. A second level uses the result of this pre-classification to operate a discrimination by the construction of fuzzy decision trees. The two levels are then combined to deduce from these two kind of informations (intrinsic and discriminant) the final classification. The evaluation of the system was done on on-line handwritten digit recognition tasks. The results show the interest of the hybrid and hierarchic combination and the good performances in comparison with a classifier dedicated to on-line handwriting recognition and with several others classifiers (MLP, RBF, fuzzy decision trees, SVM).
Complete list of metadatas

https://hal.archives-ouvertes.fr/hal-01191742
Contributor : Nicolas Ragot <>
Submitted on : Wednesday, September 2, 2015 - 2:29:37 PM
Last modification on : Tuesday, July 2, 2019 - 4:02:03 PM

Identifiers

  • HAL Id : hal-01191742, version 1

Citation

Nicolas Ragot, Eric Anquetil. Combinaison hiérarchique de systèmes d'inférence floue : application à la reconnaissance en-ligne de chiffres manuscrits. Conférence Internationale Francophone sur l'Écrit et le Document (CIFED'02), Oct 2002, Hammamet, Tunisie. pp.305-314. ⟨hal-01191742⟩

Share

Metrics

Record views

272