Skip to Main content Skip to Navigation
Theses

Fonctions de Croyance et Indexation Multimodale
Application à l'Identification de Personnes dans des Albums

Abstract : This thesis presents an information fusion strategy for digital photograph semi-automatic indexing using people's identities. The final goal of this work is to help users to recognize the identities of people in photo. Our fusion process is named -FUSINDEX-. It has been integrated in a webbased management application which allows user to share photos with a community and add textual and vocal comments to a selected photo. Fusindex combines the information issued from multiple modalities such
as image, text, speech, social context, date and geographic position. We used several indexing engines. Each one of these engines analysis a single modality: the image itself, textual comments and vocal comments. But in this paper we focalize essentially on engines which analyze the
image. The algorithms of Fusindex are based on the evidence theory of Dempster-Shafer in order to manage, as well as possible the imprecision and the uncertainty of the results obtained by the indexing engines.
Complete list of metadata

Cited literature [87 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00232806
Contributor : Said Kharbouche <>
Submitted on : Friday, February 1, 2008 - 5:59:14 PM
Last modification on : Tuesday, February 5, 2019 - 11:44:21 AM
Long-term archiving on: : Monday, May 3, 2010 - 8:05:22 PM

Identifiers

  • HAL Id : tel-00232806, version 1

Citation

Said Kharbouche. Fonctions de Croyance et Indexation Multimodale
Application à l'Identification de Personnes dans des Albums. Interface homme-machine [cs.HC]. Université de Rouen, 2006. Français. ⟨tel-00232806⟩

Share

Metrics

Record views

269

Files downloads

328