December 3, 2007
Interschool Lab, CEPSR 750
Hosted by: Prof. Shih-Fu Chang
Speaker: Dr. Apostol (Paul) Natsev, IBM T. J. Watson Research Center
Digital video consumption has skyrocketed in recent years and is rapidly becoming commonplace in many parts of our lives -- from the way we entertain and inform ourselves to the way we communicate, socialize, or learn. With the tremendous growth of video come great opportunities but also greater expectations and challenges. Users expect video to be easily searchable but technology has unfortunately not kept pace, and relevant video is difficult to find in the increasing deluge of content. Traditional approaches of video indexing based on manual tagging, textual metadata, and link analysis are coarse-grained and inadequate. Content-based indexing approaches are sometimes more effective but are plagued by scalability and usability problems, and are usually stuck in the research lab.
In this talk, we will present the current state of video search, and will review existing approaches for multimodal video indexing and search. Emphasis will be given on a new promising direction of research, called concept-based video retrieval, which aims to boost both the effectiveness and usability of video search. We will describe techniques that leverage the computer's ability to effectively analyze visual features of video and apply statistical machine learning techniques to classify and label video scenes automatically. We will also describe methods that leverage such automatically generated labels to improve the quality of video indexing and search. All approaches will be presented and demonstrated in the context of a state-of-art video analysis and retrieval system developed at IBM Research.
Dr. Apostol (Paul) Natsev is a Research Staff Member and Manager of the Multimedia Research Department at the IBM T. J. Watson Research Center. Dr. Natsev received his M.S. (1997) and Ph.D. (2001) degrees in Computer Science from Duke University, and joined IBM Research in 2001. His primary research interests are in the areas of multimedia semantic indexing and search, multimedia understanding and machine learning, as well as multimedia databases and query optimization.
Dr. Natsev is a founding member of the IBM Research Multimedia Analysis and Retrieval project (MARVEL), which was awarded the Wall Street Journal Innovation Award (Multimedia category) in 2004, and an IBM Outstanding Technical Accomplishment Award in 2005. He is an active participant in the NIST TREC Video Retrieval (TRECVID) evaluation, and leads the IBM video search team, which has achieved excellent performance in TRECVID several years in a row.