Learning a classifier with very few examples: knowledg based and analogy generation of new exemples for character recognition.

Sabri Bayoudh 1 Harold Mouch?e 2 Laurent Miclet 1 Eric Anquetil 2
1 CORDIAL - Human-machine spoken dialogue
IRISA - Institut de Recherche en Informatique et Systèmes Aléatoires, INRIA Rennes, ENSSAT - École Nationale Supérieure des Sciences Appliquées et de Technologie
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 : This paper is basically concerned with a practical problem: the on-the-fly quick learning of handwritten character recognition systems. More generally, it explores the problem of generating new learning examples, especially from very scarce (2 to 5 per class) original learning data. It presents two different methods. The first one is based on applying distortions on original characters using knowledge on handwriting properties like speed, curvature etc. The second one consists in generation based on the notion of analogical dissimilarity which quantifies the analogical relation “A is to B almost as C is to D”. We give an algorithm to compute the k-least dissimilar objects D, hence generating k new objects from three examples A, B and C. Finally, we experimentally prove on 12 writers the efficiency of both methods, especially when used in conjunction.
Complete list of metadatas

https://hal.inria.fr/inria-00300717
Contributor : Harold Mouchère <>
Submitted on : Friday, July 18, 2008 - 6:16:44 PM
Last modification on : Monday, January 14, 2019 - 10:00:13 AM

Links full text

Identifiers

Citation

Sabri Bayoudh, Harold Mouch?e, Laurent Miclet, Eric Anquetil. Learning a classifier with very few examples: knowledg based and analogy generation of new exemples for character recognition.. European Conference on Machine Learning, Sep 2007, Warsaw, Poland. ⟨10.1007/978-3-540-74958-5_49⟩. ⟨inria-00300717⟩

Share

Metrics

Record views

195