Recherche approximative de plus proches voisins avec contrôle probabiliste de la précision ; application à la recherche d'images par le contenu

Sid-Ahmed Berrani 1, 2
2 TEXMEX - Multimedia content-based indexing
IRISA - Institut de Recherche en Informatique et Systèmes Aléatoires, Inria Rennes – Bretagne Atlantique
Abstract : This thesis is concerned with content-based image retrieval (CBIR) systems. These systems retrieve the images that are the most similar to a query one within an image database. This requires describing automatically the images such that the visual similarity between two images becomes proportional to the distance between their descriptors. Retrieving the most similar images of the query images is therefore achieved by searching for the nearest-neighbors of the descriptor of the query image. This search mode is however inherently expensive when the database is very large and even when multidimensional indexing techniques are used. The dimensionality curse problem makes these techniques inefficient in dimensional spaces. It is possible, however, tor educe the retrieval cost by searching for approximate neighbors of the query descriptors instead of searching for the exact result. In this thesis, this principle is used and a new approach for performing efficient approximate nearest-neighbor searches in high dimensional databases is proposed. This approach allows a fine and intuitive control over the precision of the search by setting the maximum probability to miss one of the exact nearest neighbors. In addition, this method is used to perform similarity searches within a CBIR system for image copy identification. The idea is to evaluate the method within a real system. This evaluation shows that it is efficient, reliable and its imprecision has only a little impact on the final results
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00532854
Contributor : Patrick Gros <>
Submitted on : Thursday, November 4, 2010 - 3:36:58 PM
Last modification on : Friday, November 16, 2018 - 1:30:28 AM
Long-term archiving on : Saturday, February 5, 2011 - 2:30:25 AM

Identifiers

  • HAL Id : tel-00532854, version 1

Citation

Sid-Ahmed Berrani. Recherche approximative de plus proches voisins avec contrôle probabiliste de la précision ; application à la recherche d'images par le contenu. Interface homme-machine [cs.HC]. Université Rennes 1, 2004. Français. ⟨tel-00532854⟩

Share

Metrics

Record views

522

Files downloads

1155