Wednesday, February 19, 2003
at 5:30PM

Toward more intelligent multimedia information systems - CueVideo

CueVideo deals with fully automated indexing and cross-indexing of variety of media such as video and audio from presentations and talks. Its intended target is fully automated creation of WWW multimedia repositories of such presentations allowing fast browse as well as textual queries. In CueVideo we studied two challenging issues: how to automatically create video summaries that allow fast browse and playback of videos; and issues related to using standard speech recognition SW like IBM ViaVoice speech recognition for automated semantic indexing into audio portion of the video. For fast video browsing, we showed by empirical user study that humans could
"absorb" and comprehend videos at rates of 50 - 70% faster than normal speed, thus greatly increasing efficacy of browsing. In use of speech recognition we showed that it can produce reasonably good results (precision and recall in the range of 0.7 - 0.8) in cases where audio quality is high but that more work needs to be done in terms of tuning up its language model and vocabulary to the application domain. We will show experimental results on the database of 6 hours of indexed video compared with ground truth. This work was done while author was in IBM Almaden Research Center.

We will end up the talk by briefly making the case for use of more computer-based "intelligence" in the design, usage and management of large computer systems, which is a new research direction that the author advocates. Its applications are for example in autonomic systems management, adaptive user interfaces, and more intelligent information gathering from large heterogeneous data sources or sensor-intensive systems.

Dr. D. Petkovic obtained his Ph.D. at UC Irvine, in the area of biomedical image processing. He spent over 15 years at IBM Almaden Research Center as a scientist and in various management roles. His contributions ranged from use of computer vision for inspection, to multimedia databases and digital libraries. He is founder of well-known IBM's QBIC (query by image content) project, which significantly influenced content-based retrieval field. Dr. Petkovic received numerous IBM awards for his work and became IEEE Fellow for leadership in content-based retrieval area. Dr. Petkovic also managed and participated in several other projects while in IBM, among them User Ergonomics Research (involved in IBM's TrackPoint pointing
device), and Foundations of Massively Parallel Computing.

In last few years Dr. Petkovic was having various technical management roles in Silicon Valley startups, the latest (VMware) involving virtual computing on Intel Platform. Some of the products Dr. Petkovic helped build won numerous awards and are widely used. Currently, Dr. Petkovic is teaching at Santa Clara University.