Skip to Main content Skip to Navigation
Theses

Méthodes d'apprentissage pour la recherche de catégories dans des bases d'images

Philippe-Henri Gosselin 1
1 MIDI - Multimedia Indexation and Data Integration
ETIS - UMR 8051 - Equipes Traitement de l'Information et Systèmes
Abstract : The decreasing cost of stockage and availability of high quality numeric media allow today the building of huge image databases in various scopes. These databases are often under exploited, because of the lack of tools to access the information they contain. In this thesis, we propose a set of learning techniques for image category search, gathered in the RETIN 2 system. Concerning indexing, we propose a quantization method for a very large number of vectors in order to build the image indexes as histograms, and also a kernel approach of similarity. Concerning interactive search, we study several classification methods, and present active learning which aims at selecting the pictures the user should label. We propose several methods in an active learning architecture which deal with the particular characteristics of the CBIR context. At last, we study the problem of the re-use of labels users have given during retrieval sessions. We present the weakly aspect of this learning problem in our context. We propose two methods which optimize the similarity between images in the scope of kernel functions. The system is evaluated on a generalist image database of several thousands of images.
Complete list of metadata

Cited literature [111 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00619222
Contributor : Philippe-Henri Gosselin <>
Submitted on : Monday, September 5, 2011 - 6:07:52 PM
Last modification on : Monday, January 25, 2021 - 3:16:02 PM
Long-term archiving on: : Tuesday, December 6, 2011 - 2:28:41 AM

Identifiers

  • HAL Id : tel-00619222, version 1

Citation

Philippe-Henri Gosselin. Méthodes d'apprentissage pour la recherche de catégories dans des bases d'images. Interface homme-machine [cs.HC]. Université de Cergy Pontoise, 2005. Français. ⟨tel-00619222⟩

Share

Metrics

Record views

512

Files downloads

348