Briefs

InformationWeek (08/30/04) No. 1003, P. 44; Ricadela, Aaron

TV news broadcasters, PC users, and intelligence analysts are just some of the people who could benefit from video search technologies being developed by IBM, Microsoft, and academic researchers. IBM recently demonstrated Marvel, a prototype computer system that employs statistical methods to establish relationships between video footage elements such as color, sound, shapes, and patterns, and label the footage appropriately so that users can find individual shots. Project coordinator John Smith's intelligent information management group has authored algorithms that can recognize 140 concepts culled from an archive of ABC News and CNN broadcasts held by the University of Pennsylvania, and the team is planning a November report on their work to boost Marvel's accuracy by combining concepts.

Smith's group is also collaborating with Columbia University's digital video multimedia lab to integrate computer vision and image understanding with machine learning strategies so that news footage from American and foreign broadcasters can be mined for related topics. Columbia University electrical engineering professor Shih-Fu Chang observes that video searching is complicated by video's lack of textual or graphical structure. Meanwhile, Carnegie Mellon University's Informedia initiative is developing a technique to answer video search queries by evaluating elements such as shapes, text, and colors. And Microsoft Research has developed a system that enables users to view all relevant shots in a home movie by moving their mouse pointers over objects they wish to see.

Here is the rest of the story.