next up previous
Next: Introduction

Enhancement of Textual Images Classification using their Global and Local Visual Contents

Sabrina Tollari - Hervé Glotin - Jacques Le Maitre

Laboratoire SIS - Equipe informatique,
Université de Toulon et du Var,
Bâtiment R, BP 20132,
F-83957 La Garde cedex, France

tollari@univ-tln.fr glotin@univ-tln.fr lemaitre@univ-tln.fr

PDF PS PPT BIB

Abstract:

This paper deals with the existence of a dependance between the textual indexation of an image (a set of keywords) and its visual indexation (color and shape attributes). This experience has been realized on a corpus of news photos manually indexed by keywords extracted from a hierarchically structured thesaurus. First, a reference classification of these photos has been constructed from their textual indexation (regarded as relevant), then textual and visual features characterizing these classes have been constructed. Finally, they have been used to evaluate performances of a content-based image retrieval combining textual and visual descriptions. Results of the visuo-textual classification show an improvement of 54% against classification using only textual information.
Keywords Information retrieval, content-based image retrieval(CBIR), visuo-textual fusion, vectorial model





Tollari Sabrina 2003-08-28