April 1th, 2002

Title : A Variational Framework for Simultaneous Segmentation
and Registration, Shape Averaging, and Tracking of Deforming Objects

Anthony Yezzi
Georgia Institute of Technology

In this talk we begin by presenting a technique to simultaneous segmentation and registration using the variational framework of active contours. This work was motivated by the desire to segment and register individual vertebra from MR and CT spine images. Next, we extend this model to shapes which are nonequivalent (with respect to the group through which we wish to register them) by intorducing a notion of "average shape". We then show how this may be formulated into a robust approach to active contour based video tracking in which we attempt to "approximately segment" multiple frames at a time rather than "accurately segment" each instantaneous frame.