VASC Seminar Announcement
            =========================

Date: Monday, 5/21/01
Time: 3:30-4:50
Place: NSH 3002

Speaker: Jonas August, Center for Computational Vision and Control, Yale University
Title: Taming Skeleton Instabilities with Robust Curve Inference

Consider the problem of comparing one shape, such as a hammer, to a variety
of very different shapes, such as elephants, planes, and people.
Hierarchical descriptions of shapes, based on skeletons or medial axes for
example, can support such shape comparisons by revealing each shape's part
structure and part relevance.  Unfortunately, the skeleton has long
been viewed with suspicion due to its sensitivity to noise or gaps in the
boundary.  We first address the "causes" of this sensitivity by studying
the reliable inference of boundaries despite noise and the grouping of
contours fragmented by occlusion.  After, we study the nature of this
sensitivity by analyzing the stability of the skeleton under boundary
motions.

To infer object boundaries from an image, we propose a random field model
of an ideal image of contours, where each contour is modeled using a Markov
process.  We show that this "curve indicator random field" is non-Gaussian
and yet we can still provide a tractable set of joint moments.  These
moments allow us to derive nonlinear Bayesian filters for enhancing
contours in noisy images.  Even such filtered contours can still be
fragmented due to occlusion, and so we propose the principle that those
fragments should be grouped whose fragmentation could have arisen from a
shared, simple occluder.  We introduce the gap skeleton both as a
representation of this "virtual" occluder and as a means for linking the
contour fragments into groups.

Finally, we derive the effects of perturbations of the shape on the
skeleton.  In particular, we characterize how smooth points of the skeleton
evolve under a general boundary evolution, with the corollary that, when
the boundary is smoothed by a geometric heat equation, the skeleton evolves
according to a related geometric heat equation.  The surprise is that,
while certain aspects of the skeleton simplify, as one would expect, others
can behave wildly, including the creation of new skeleton branches.
Fortunately such sections can be flagged as "ligature," or those portions of
the skeleton related to shape concavities.  Ligature is now being used to
improve the performance of a generic object recognition system.