Hiérarchies sémantiques pour l'annotation multifacette d'images

Anne-Marie Tousch 1, 2
2 IMAGINE [Marne-la-Vallée]
LIGM - Laboratoire d'Informatique Gaspard-Monge, CSTB - Centre Scientifique et Technique du Bâtiment, ENPC - École des Ponts ParisTech
Abstract : In this thesis, we address the problem of automatic image annotation. For a more flexible system, we build multi-faceted annotations organized in a semantic hierarchy. Thus, an annotation is defined by a set of multilabels coupled with confidence levels. A tradeoff between reliability and semantic precision allows greater flexibility. The proposed algorithm proceeds in two stages. First, informative image features are extracted. Second, normalized probabilities are computed on a set of multilabels. Both rely on statistical learning machines. We evaluate the approach on two datasets : a set of car images and a generic database, Caltech-101. Results show different behaviour depending on the data, suggesting that the vocabulary structure is useful at different stages of the algorithm.
Document type :
Theses
Complete list of metadatas

https://pastel.archives-ouvertes.fr/pastel-00555122
Contributor : Vivien Girard <>
Submitted on : Wednesday, January 12, 2011 - 1:53:09 PM
Last modification on : Thursday, July 5, 2018 - 2:29:02 PM
Long-term archiving on : Wednesday, April 13, 2011 - 2:49:57 AM

Identifiers

  • HAL Id : pastel-00555122, version 1

Citation

Anne-Marie Tousch. Hiérarchies sémantiques pour l'annotation multifacette d'images. Traitement des images [eess.IV]. Ecole des Ponts ParisTech, 2010. Français. ⟨NNT : 2010ENPC1002⟩. ⟨pastel-00555122⟩

Share

Metrics

Record views

640

Files downloads

390