Seminars/Colloquia/Invited Talks

Seminars

Gianfranco Doretto

Image Descriptors for Video Surveillance Applications

PLACE:	Clark 314
EVENT:	CIS Seminar
DATE:	April 24, 2007
TIME:	1:00 - 2:00 PM

Abstract

Modeling object appearance in order to compute the similarity of certain image regions is a fundamental problem in object identification. In order to achieve the highest recognition rates, one would have to design models that capture the distinctiveness of individual objects within a given category while maintaining invariance with respect to illumination and pose changes, as well as object deformation. Since we target video surveillance applications, the computational complexity of the model becomes crucial in order to guarantee the desirable real-time performances, and the challenging problem is to strike a balance between the distinctiveness, invariance, and computational complexity of the model.
In this talk we introduce the concept of shape and appearance context, which dramatically improves the distinctiveness of appearance models while maintaining a high degree of invariance. The approach uses a local appearance descriptor and partitions the image region into his constituent parts using a modified shape context algorithm. The shape and appearance context is the descriptor that captures the spatial distribution of the appearance relative to each of the parts. We will introduce a generalization of the popular integral image and integral histogram "tricks", and we will show how they allow a remarkable computational complexity decrease, enabling a real-time computation of the shape and appearance context. Finally, we will show the ability of the framework to match the identity of people from query images of them.

Brief Biography:

Gianfranco Doretto is a research scientist at General Electric Global Research. He received the Ph.D. and M.S. degrees in Computer Science from the University of California, Los Angeles, in 2002 and 2005 respectively, and a D.Eng. degree in Electronics Engineering (with highest honors) from the University of Padua, Italy, in 1998. His research interests span several areas of computer vision, with a current focus on statistical video modeling for aerial and ground surveillance applications. He has authored more than twenty publications, and he is a member of the IEEE, the IEEE Computer Society, and Sigma Xi.