Section: Contracts and Grants with Industry
European Network of Excellence MUSCLE: Multimedia Understanding through Semantics, Computation, and Learning
Duration: 4 years, starting in April 2004. 42 partners. Prime: ERCIM, scientific coordinator: Nozha Boujemaa, INRIA.
This project aims at developing the collaboration in the domain of automatic multimedia document analysis, in particular to be able to handle and exploit their meaning. The project is thus concerned by all content-based analysis tools available for every media (text, sound and speech, image and video), but also by the techniques which allow us to combine the information extracted from each media, and by the common techniques needed to handle such data (optimization, classification, intensive computation).
TexMex is mainly active in the WP6 (multimodal analysis) through the work of M. Delakis (see Section 6.3.1 ).
European Integrated Project aceMedia: Integrating Knowlege, Semantics and Content for User-Centered Intelligent Media Services
Duration: 48 months, started in January 2004. 15 partners. Prime: Motorola Ltd.
The goal of this project is to encode multimedia document for their diffusion on networks like Internet, telecommunication networks or broadcasting systems. This new encoding scheme is based on autonomous entities called ACEs (standing for Autonomous Content Entity). Each entity is made of data, related metadata and an intelligence layer.
ACEs are dedicated to storing, retrieving and communicating documents in an efficient and autonomous way. It supports self-organization, self-annotation and self-adaptation according to the current user's preferences and devices. Additional embedded mechanisms are semantic detection, fast retrieval and relevance feedback.
TexMex team is responsible for the methods related to indexing, intelligent search and fast retrieval of ACE documents. The ACE documents will be described by both conceptual and content-based descriptors. Our algorithms compute the list of the most similar documents according to a query, by matching their numerical descriptors. Currently, the provided methods are the optimised exhaustive search and a based-on kd-trees method. Other methods will be evaluated.