Contributions to image retrieval in the wavelet transform domain

Résumé

This thesis addresses the problem of images indexing and retrieval in the wavelet transform domain. In particular, two major issues are considered: the indexing of stereo images and the impact of quantization in still image retrieval schemes. In the first part, we propose novel retrieval approaches devoted to stereo images which integrate the disparity information with the visual contents of stereo images. In the first strategy, the two views are processed separately through a univariate model. An appropriate bivariate model is employed to exploit the cross-view dependencies in the second method. In the third strategy, we resort to a multivariate model to further capture the spatial dependencies of wavelet subbands. In the second part, different strategies are designed to improve the drop of retrieval performances resulting from the quantization of database or query images. First, we propose to operate on the quantized coefficients by applying a processing step that aims at reducing the mismatch between the bitrates of the model and the query images. As an alternative, we propose to recover the statistical parameters of original wavelet coefficients directly from the quantized ones. Then, we investigate different quantization schemes and we exploit inherent properties of each one in order to design an efficient retrieval strategy.

Citer ce document

Accès au document

Ce document est hébergé sur son dépôt institutionnel d'origine.

Auteur(s)

Statistiques

Consultations : 2

Téléchargements : 0