COLLOQUIUM
Department of Computer Science and Engineering
University of South Carolina
 
Robust Feature Analysis: From Algorithms to Biology

Toshiro Kubota
Department of Computer Science and Engineering
University of South Carolina

Date: November 22, 2002 (Friday)
Time:  3:30-4:30PM
Place: Swearingen 1C01 (Amoco Hall)


Abstract

Automated image analysis and interpretation is a very hard problem for
several reasons.  First, it is fundamentally hard, as an automated
system has to recover 3D information from 2D projections.  Second,
there are so many data points in an image that it is a super-
high-dimensional data analysis problem. Third, since we do it very
well, its difficulty is often underestimated.  Fourth, its goals are
often so descriptive that it lacks an appropriate numerical measure
for performance analysis and comparison.  In this talk, I start by
giving a brief overview of computer vision research and trying to
convince the audience that it is a very challenging but intriguing
subject.  I argue that many vision tasks can be improved if we take
contextual information from the very beginning of data processing.  I
propose a computational framework that incorporates contextual
information and describe some algorithms for various vision problems.
Finally, I show a possible link between the proposed framework and
adaptive properties of biological neural systems.



Toshiro Kubota received the BS degree in Instrumentation Engineering
from Keio University, Japan in 1988, the MS degree in Electrical
Engineering from Georgia Institute of Technology in 1989 and the Ph.D.
degree in Electrical and Computer Engineering from Georgia Institute
of Technology in 1995.  Dr. Kubota joined the University of South
Carolina in 1996 as a lecturer and research professor, and he is now
an assistant professor in the Department of Computer Science and
Engineering.  His research interests include computational vision,
image processing, neural networks and computer graphics.