TEXMEX Research team
Multimedia content-based indexing
- Leader : Patrick Gros
- Research center(s) : CRI Rennes - Bretagne Atlantique
- Field : Perception, Cognition and Interaction
- Theme : Vision, perception and multimedia interpretation
- Partner(s) : CNRS,Institut national des sciences appliquées de Rennes,Université Rennes 1
- Collaborator(s) : U. RENNES 1, CNRS, INRIA, INSA RENNES, CENTRALESUPELEC, ENS RENNES, INSTITUT MINES-TELECOM, UBS
The explosion of the quantity numerical documents multimedia generated a very strong dynamism in the field of the multimedia indexing research. However, the range of the work undertaken by the media specialists is limited by their monomedia aspect and the quantity of the documents they can handle, a few thousands of images for example, whereas the professional applications wouold require to handle much more (a few million).
Such quantities of documents set difficult problems of structuring and storage on disc, problems which are out of the expertise of the media specialists. On the other hand, database specialists who are used this kind of the problems, consider only very rudimentary techniques of document description, by lack of know-how in this field.
To remedy that, we propose the creation of a team bringing together at the same time specialists of the media and specialists of the techniques necessary to use these documents such as data bases, information retrieval or statistics. The objective of the team is thus to be at the intersection of the two following axes :
- Definition of new document descriptors for still images, video and text, definition of descriptors mixing several media and medata associated with the documents, evaluation of these descriptors on great bases of documents ; statistics for the exploration of big volumes of data, management and stategies of calculation of the metadata and the descriptors associated with the documents, data quality analysis, study of sparing strategies of exploitation (navigation, indexing, research), definition of the system and material support techniques for a fast access to these data.
- The originality of our approach comes from the simultaneous consideration of the contraints dependent on the media and the documents and of the constraints related to the exploitation of these data, wich are two aspects of the same problem. This multi-field approach must make it possible to exced the limits of the current systems and to manage finely and effectvely very significant quantities of documents.
Our work is organized in two axes of work hich we apply to the study of three problems. The axes of work are : The description of the documents multimedia : it is a question of being able automatically to calculate descriptors of the contents of a document or other medata, to check the relevance and the discriminating capacity of these descriptors at the time of research in great bases of documents the use of these descriptors for the organization and management of the bases of documents exploration and navigation, or it document retrieval : strategies of calculation, management and maintenance of the coherence of the descriptors and medata, analyses exploratory data multidimensional indexing, supports systems and materials for the systems of research.We apply these tools to three problems :
- The search for images in great bases of images ;
- The joint descriptionj text - image of documents comprising these two media ;
- The addition of semantic capacities to the textual search engines.
We first of all apply our work in the field of press and media ; archives of videos, television, photography and news services, Internet and company Intranets, the biomedical field is a large supplier of difficult data ; medical bibliographical databases, data of imagery : anatomical and functional cerebral imagery for example genomic and proteomic data.
Other applications :
Management of the robot visual mamory for motion planification analysis of the texts (relevance, sales leaflet...) for economic intelligence.
International and industrial relations
- Support for starting the team : we were supported by the ministry of research and the STIC scintific department of CNRS (JemSTIC program).
- European project : we are partner in the IST BUSMAN project, with Spanish German, and English partners. We bring in this project our expertise in data bases and indexing of image descriptors.
- National projects: we participate in many national projects with
academic and industrial
- PRIAM MEDIAWORKS project on TV archives indexing ;
- STIC Project "Indexing, transmodality and knowledge management",
- RNRT Diphonet project on indexing and watermaking techniques for copyright protection in photo agencies ;
- ACI Neurobase project on the creation of distributed databases of cerebral imagery ;
- ACIGrid GenoGRID and action bio-information inter EPST APREDG projects on the management og genomic databases
- Caderige2 project on the knowledge extraction from biologcal databases.
- Industrial contracts : we have priviledged contracts with Thomson Multimedia company on vodeo indexing techniques, and with the Thales company.