Modeling and Retrieving Audiovisual Information - A Soccer Video Retrieval System

This paper describes the results of an ongoing collaborative project between KPN Research and the Telematics Institute on multimedia information handling. The focus of the paper is the modelling and retrieval of audiovisual information. The paper presents a general framework for modeling multimedia information (ADMIRE) and discusses the application of this framework to the specific area of soccer video clips. The core of the paper is the integration of feature extraction and concept inference in a general framework for representing audio visual data. The work on feature extraction is built on existing feature extraction algorithms. The work on concept inference introduces a new approach to assigning semantics to collections of features in order to support concept-based retrieval, rather than feature-based retrieval. Finally, the paper describes our experiences with the implementation of the methods and techniques within the ADMIRE framework using a collection of commercially available tools. The latter is done by implementing a soccer video clip annotation and query tool.