Talk by Don Geman on Friday, 8 December

[Margery Ishmael]

Friday, 8 December at 2:30 pm in Ryerson (annex) 276

COARSE-TO-FINE COMPUTATIONAL VISION

Talk by Donald Geman
Department of Mathematics and Statistics
University of Massachusetts

Abstract:
I will summarize a research program in computational vision,
originally motivated by the startling efficiency of adaptive testing
in parlor games such as "twenty questions." A natural framework for
problems such as visual tracking, object detection and image retrieval
is then statistical and information-theoretic, with performance
measured by the amount of computation necessary to reach a given level
of accuracy. The recurring mathematical issue is the design and
analysis of efficient testing strategies.
First I will review some model-based work using "coarse-to-fine
packing" for measuring lesions in MRI brain scans and using entropy
coding for both road-tracking and image retrieval. Then I will
present ongoing, learning-based experiments on detecting instances
from a generic object class (e.g., a face) in natural scenes. The
efficiency criterion leads to feature selection driven by relative
likelihoods; to very sparse, discrete object representations; and to
highly coarse-to-fine processing, in both the exploration of classes
and poses, resulting in a very skewed spatial distribution of
computation.

Visitor's Host: Stuart Kurtz
====================================================
Margery Ishmael
Department of Computer Science
1100 E. 58th St.
Chicago, IL 60637

tel: 773 834-8977  fax: 773 702-8487

marge at cs.uchicago.edu