IMEDIA - Team - Research - Publications - Projects - Events - Contacts Activity report 2007

Logo INRIA

Research Themes

Indexing, navigation and browsing in large multimedia data streams are connected with several scientific domains such as image analysis, shape recognition, databases, browsing, information theory, learning and human-computer interaction.

The approach of the IMEDIA project is related to image analysis, the bottleneck of multimedia indexing. Our methodology mainly concerns: image analysis for feature space; probabilistic modelling, statistics and information theory for interactive browsing, similarity measures and matching. The methods we develop are implemented into the prototype software.

 

Image and Video Copy Detection

Alexis Joly, Julien Law-To, Nozha Boujemaa, Valerie Gouet, Olivier Buisson, Ivan Laptev, Laurent Joyeux

Muscle Copy Detection Showcase

MUSCLE VCD - Data, Live Benchmark at CIVR'07

TrecVid'08

 

Modelling, construction and structuring of the description space

Nozha Boujemaa, Valérie Gouet-Brunet, Anne Verroust-Blondet, Sabri Boughorbel, Jean-Philippe Tarel, Jean-Paul Chièze, Michel Scholl, Nizar Grira, Michel Crucianu, Hichem Houissa, Itheri Yahiaoui, Olfa Besbes, Julien Law-To, Julien Fauqueur, Mohamed Chaouch

Keywords: Image signature, visual similarity, visual indexing, image analysis, partial queries, image index management, 3D indexing , 2D shape descriptor

 

Global signature for approximate retrieval in generic image databasesGlobal signature for approximate retrieval in generic image databasesSpecific signature for face recognitionSpecific signature for face recognition

Region description schemeOriginal imageSegmented imageRegion description by ADCSAnswer to a region-based query

Partial query with local descriptionResult of the partial query with local description

 

Statistical learning and pattern recognition (more to come)

Nozha Boujemaa, Donald Geman, François Fleuret, Hichem Sahbi, Sabri Boughorbel, Nizar Grira, Michel Crucianu, Jean-Philippe Tarel, Bertrand Le Saux

Keywords: Unsupervised clustering, competitive agglomeration, SVM, hierarchy of classifiers

 

Image space categorization

Visual summary of keyframe imagesExamples of images belonging to a same cluster

Hierarchy of classifiers for face detectionIntensity of local analyses

 

Interactive search and personalization (more to come)

Nozha Boujemaa, Donald Geman, Yuchun Fang, Marin Ferecatu, Michel Crucianu, Hichem Houissa, Julien Fauqueur

Keywords: Semantic gap, subjective classification, relevance feedback, efficient visual thesaurus construction

 

Interactive search of portrait paintings by relevance feedback

Personalization by query composition

Query by composition of region categories

Results to query by composition of region categories

 

Multimedia indexing (more to come)

Nozha Boujemaa, Valérie Gouet-Brunet, François Fleuret, Marin Ferecatu, Michel Crucianu

Keywords: Hybrid image/text retrieval and annotation, image analysis

 

Automatic textual annotation (close-up estimation) after face detection and pose estimationAutomatic textual annotation (name of person) after face recognitionAutomatic textual annotation (name of person) after face recognition

 

INRIA - Rocquencourt - IMEDIA Project

© INRIA  About this site - updated June 8, 2005 - webmaster-imedia(a)inria.fr