MATHEMATICS COLLOQUIUM
Speaker: Mary Pugh
Title: Image Segmentation and Energy Dissipation.
Affiliation: University of Pennsylvania.
Date: Friday, April 13, 2001.
Place and Time: Room 101 - Love Building, 3:35-4:35 pm.
Refreshments: Room 204 - Love Building, 3:00 pm.
Abstract.
In this talk I will present some work in progress in vision
research. We consider the problem of recognizing what parts of an
image are perceived as being in the foreground. We use a variant of
the Pao-Geiger-Rubin model, which uses an energy dissipation approach
to this problem. The model is surface-based, rather than
contour-based. Specifically, the edges in the image are not viewed as
isolated contours, but are viewed as bounding a surface. Each local
edge has a local hypothesis; for example, a north-south edge might
think "the region immediately to the left of me is part of the
figure". The model then uses energy dissipation methods to seek
assignments of local hypotheses that are mutually agreeable, yielding
a segmentation of the image that might be perceived. We test the model
on various images to address questions like: does the model "perceive"
objects to be in the foreground (the way we do)? how does it perform
on optical illusions that viewers report to have two different
segmentations?
This is joint work with Nava Rubin of the Center for Neural Science,
NYU and Davi Geiger of the Courant Institute, NYU. I thank Anita
Disney (CNS, NYU), Dave McLaughlin (Courant, NYU) and Bob Shapley
(CNS, NYU) for useful discussions.
|