ERCIM News No.35 - October 1998
Digital Media Warehouse Systems - Querying a National Multimedia Database
by Martin Kersten
CWI has started research aiming at the advance of content-based retrieval techniques in a large multimedia database. It is a cooperative effort under the umbrella of the recently founded national Telematics Institute, in which several knowledge institutions and industries participate. Other partners are the University of Twente, KPN and Syllogic. CWI also manages the project.
Optimal exploitation of the information contained in such a database, or Digital Media Warehouse (DMW), requires strategic and fundamental research in several fields of telematics. The project is focused on the system architectural needs for querying a DMW, in the context of video and audio service directories.
Searching information in a DMW is often not fixed in its form, but requires browsing interaction and proximity queries to find a satisfactory solution. But how can the owner of a DMW efficiently organize and disclose its content if he can not anticipate the storage, communication and query characteristics to optimize its architecture? In particular, a large portion of the database resides in semi-structured archives (Web-pages).Whereas (manual) indexing cannot be handled economically, automatic indexing multimedia is still an unsolved problem.
In areas such as computer vision, image processing, multi-media authoring, and information retrieval, the aim is to develop feature detectors as building blocks for content-based indexing, interpretation, and classification. The novelty, and prime objective of the present project, lies in developing an adaptive system architecture allowing growth and easy access of multi-terabyte databases stored globally. This encompasses a reference framework for feature detectors extracting relevant feature values from different kinds of multimedia objects, a database architecture for their storage, and query techniques for semi-structured data. The DMW project is unique in that it combines these pieces for support of content-based retrieval in an integrated and scalable solution.
The DMW research aligns with the national project Amis, directed at multimedia indexing and content-based search, in which CWI develops the Acoi database containing a large collection of multimedia objects, that acts as a national experimentation platform, and with the national HPCN projects Jera (large-scale webservers) and IMPACT (high-performance database architectures).
Acois initial release is geared at provision of one million still images, hundreds of video sequences, and thousands of audio tracks. This large multimedia database is subsequently indexed using feature extraction algorithms published in the literature and developed by the partners in the various affiliated research projects. The prime demonstrator is the Acoi search engine. The database is built under Monet, our extensible database system with demonstrated high performance. The hardware platform is an SGI O2000 computer with 16 CPUs at 250Mhz, 8 Gbyte ram, and 150 Gbyte disk space. The system is hooked into the communication infrastructure using 4 x 150 Mbyte ATM links.
The envisioned DMW architecture for experimentation consists of three layers of activity:
- storage and WWW access, built around the Monet database system
- query and feature detection, comprising two components dealing - roughly speaking - with DMW input and output - applications to highlight and exploit various aspects of the platform.
More information on website: http://www.cwi.nl/~acoi/DMW/index.html
Martin Kersten - CWI / University of Amsterdam
Tel:+31 20 592 4066