A Semantic Views Model For Audiovisual Indexing and Retrieval


PhD thesis N°2738, EPFL, Lausanne, March 2003

2001-2003 / Author: Nastaran Fatemi


Abstract: This PhD work presents an innovative semantic model for an audiovisual information retrieval system which better covers the user’s requirements in an integrated TV news production environment. This model exploits an interchangeable description format (NN Schema) that we proposed in order to integrate a rich set of required TV news information. In this work, we firstly propose a semantic model for the retrieval of audiovisual information, called the Semantic Views Model. This model takes into account various users’ viewpoints from which they search for audiovisual content in the context of their activities. We propose two different modes to retrieve audiovisual content: querying and browsing and strategies to integrate these two modes in order to improve the interactive retrieval. To facilitate querying, we propose a high-level query language following the Semantic Views Model, called Semantic Views Query Language (SVQL). Moreover to allow browsing and playing, we define the Semantic Views Hypermedia Model. We secondly propose NN Schema, a TV news description schema based on MPEG-7, which allows the representation of TV news information in an environment that integrates the three processes of production, archiving and retrieval. The use of NN Schema improves all the three so-called processes by allowing an effective information exchange between them and also with external applications. To validate our proposed methodology, we set up the COALA experimental project. COALA provides various tools for both indexing and retrieval of audiovisual programs at the TSR broadcasting company. The design of these tools is based on the application of the Semantic Views Foma1 Model and the WN Schema specification. These principles succeeded in overcoming the challenges related to the complexity of the users’ requirements and the richness and diversity of information resources.