Un nouveau système indépendant de rejet multi-seuils pour la reconnaissance de mots manuscrits

Laurent Guichard 1, 2 Alejandro Toselli 2, 1 Bertrand Coüasnon 3, 2
2 IMADOC - Interprétation et Reconnaissance d’Images et de Documents
UR1 - Université de Rennes 1, INSA Rennes - Institut National des Sciences Appliquées - Rennes, CNRS - Centre National de la Recherche Scientifique : UMR6074
Abstract : In the field of isolated handwritten word recognition, the development of highly effective verification systems to reject words that do not belong to the lexicon or present ambiguities is still an active research topic. In this paper, a novel verification architecture based on support vector machine scoring and on a new algorithm for tuning multiple reject class-dependent thresholds is presented. In essence, a set of support vector machines appended to a standard HMM based recognition system provides class-dependant confidence measures employed by the verification mechanism to accept or reject the recognized hypotheses. In order to tune these class-dependent rejection thresholds, an algorithm based on dynamic programming is proposed which focus on maximizing the recognition rate for a given imposed error rate. Preliminary reported results of experiments carried out on Techno-Vision program RIMES database show that this approach outperforms state of the art rejection methods.
Complete list of metadatas

Cited literature [18 references]  Display  Hide  Download

https://hal.inria.fr/hal-01197490
Contributor : Bertrand Coüasnon <>
Submitted on : Friday, September 11, 2015 - 6:33:41 PM
Last modification on : Friday, November 16, 2018 - 1:21:53 AM
Long-term archiving on : Tuesday, December 29, 2015 - 12:47:01 AM

File

rfia2010guichard.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01197490, version 1

Citation

Laurent Guichard, Alejandro Toselli, Bertrand Coüasnon. Un nouveau système indépendant de rejet multi-seuils pour la reconnaissance de mots manuscrits. RFIA, Jan 2010, Caen, France. pp.471-478. ⟨hal-01197490⟩

Share

Metrics

Record views

176

Files downloads

111