Population and query interface for a content-based video database

Abstract

In this paper we describe the first full implementation of a content-based indexing and retrieval system for MPEG-2 and MPEG-4 videos. We consider a video as a collection of spatiotemporal segments called video objects; each video object is a sequence of video object planes. A set of representative video object planes is used to index each video object. During the database population, the operator, using a semi-automatic outlining tool we developed, manually selects video objects and insert some semantical information. Low-level visual features like color, texture, motion and geometry are automatically computed. The system has been implemented on a commercial relational DBMS and is based on the client server paradigm. The database population client is a stand-alone Windows application; the querying of the system is performed with a web-based client. To test the system we indexed a number of videos taken from Italian soccer championship games. A few examples of queries in this particular domain are reported and results appear very promising. © Springer-Verlag Berlin Heidelberg 2002.