Allograph Based Writer Adaptation for Handwritten Character Recognition

Kumar Chellapilla 1 Patrice Simard 1 Ahmad Abdulkader 1
1 Microsoft Research [Redmond]
Abstract : Writer adaptation is the process of converting a generic (writer-independent) handwriting recognizer into a personalized (writer-dependent) recognizer with improved accuracy for a particular user. While training the generic recognizer uses large amounts of data from several writers, the adaptation process uses only a few samples from a single user. In this paper we present a) an automatic approach for identifying allographs (character shapes/styles) from handwritten characters through clustering, b) a novel architecture for a personalizable recognizer that utilizes allograph information, and c) a kernel based approach for personalizing the recognizer. Using the new approach, personalization results with twenty one users indicate that handwritten single character recognition errors can be reduced by over 24% (or 41%) using as few as 5 (or 15) samples.
Keywords : Writer adaptation personalization allographs handwriting character recognition
Type de document :
Communication dans un congrès
Guy Lorette. Tenth International Workshop on Frontiers in Handwriting Recognition, Oct 2006, La Baule (France), Suvisoft, 2006
Liste complète des métadonnées

Littérature citée [11 références]  Voir  Masquer  Télécharger
https://hal.inria.fr/inria-00112632
Contributeur : Anne Jaigu <>
Soumis le : jeudi 9 novembre 2006 - 13:52:44
Dernière modification le : jeudi 9 novembre 2006 - 16:43:55
Document(s) archivé(s) le : mardi 6 avril 2010 - 21:59:55

Fichier

cr1039113319113.pdf

Identifiants

  • HAL Id : inria-00112632, version 1

Collections

IWFHR10

Citation

Kumar Chellapilla, Patrice Simard, Ahmad Abdulkader. Allograph Based Writer Adaptation for Handwritten Character Recognition. Guy Lorette. Tenth International Workshop on Frontiers in Handwriting Recognition, Oct 2006, La Baule (France), Suvisoft, 2006. 〈inria-00112632〉

Exporter

BibTeX TEI DC DCterms EndNote

Partager

Métriques

Consultations de la notice

131

Téléchargements de fichiers

177

Contact


archive-ouverte@inria.fr