         
|
GRASP Lab Seminar 2003-2004
April 30, 11:00 AM, Levine Hall 307, hosted by Jianbo Shi.
James Rehg
Georgia Institute of Technology
Learning a Rare Event Detection Cascade by Direct Feature Selection
Abstract: Face detection is a canonical example of a rare event detection problem, in which target patterns occur with much lower frequency than non-targets. Out of millions of face-sized windows in an input image, for example, only a few will typically contain a face. Viola and Jones recently proposed a cascade architecture for face detection which successfully addresses the rare event nature of the task. A central part of their method is a feature selection algorithm based on AdaBoost. We present a novel cascade learning algorithm based on forward feature selection which is two orders of magnitude faster than the Viola-Jones approach and yields classifiers of similar quality. This faster method could be used for more demanding classification tasks, such as on-line learning or searching the space of classifier structures. In addition, we describe an improved training criterion, based on the idea of an indifference curve, and a perturbation bias technique which results in better classification performance.
This is joint work with Jianxin Wu, Matthew D. Mullin, Jie Sun and Aaron Bobick.Biography: James M. Rehg is an Associate Professor in the College of Computing at the Georgia Institute of Technology. He is a member of the GVU Center and co-directs the Computational Perception Lab along with professors Aaron Bobick, Frank Dellaert, and Irfan Essa. Dr. Rehg joined the faculty in August, 2001. In Fall 2001 he received an NSF Career Award.
From 1995 - 2001 he worked at the Cambridge Research Laboratory in Cambridge, MA, which was initially owned by the Digital Equipment Corporation and was later acquired by Compaq Computer Corporation. From 1997 - 2001, Dr. Rehg led a research group working in the areas of computer vision, machine learning, and HCI. He received his Ph.D. in Electrical and Computer Engineering from Carnegie Mellon University in 1995. His current research interests include computer vision and machine learning and their application to HCI, computer graphics, and computer systems. Previous research topics include the DigitEyes hand tracking system (first real-time visual tracking results for complex articulated motion), the Smart Kiosk (an early example of vision-based HCI), and Stampede (a cluster parallel programming system for streaming data). Dr. Rehg is active in the program committees of the International Conference on Computer Vision (ICCV), the European Conference on Computer Vision (ECCV), and the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). He is currently serving on the editorial board of the International Journal of Computer Vision. He is a member of the IEEE and ACM.
