CS545I, Feb 16

Prof. Carlo Tomasi, Stanford Computer Science Dept.

Seminar abstract

This research concerns a content-based image retrieval system, where queries are image-like objects. In order to achieve query times that are better than linear in the number of images, images are processed at entry time to yield a large number of indices. Since these indices live in a metric space, queries can proceed in sublinear time by using appropriate search structures. We are currently investigating two types of indices: texture and shape. This lecture will discuss the query approach and the issues involved, the nature of our texture and shape indices, and the main elements of the query system. Preliminary results from two separate demonstration systems will be presented.