CS Colloquium on December 18, 2003 at 2pm

Title: Moving morphable models toward photorealism

Speaker: Matthew Brand
         Senior Research Scientist
         Mitsubishi Electronic Research Labs
         Cambridge, MA
         http://www.merl.com/people/brand/

Date: December 18, 2003 (Thursday)
Time: 2pm
Place: MCS 135, 111 Cummington Street

Abstract:

Morphable models have been moderately successful in computer vision and
highly successful in post-production graphics.  The difference is one of
economy: An artist in an FX studio will construct a head model with
hundreds of morph targets to cover the range of desired deformations for a
single character, whereas in computer vision one desires a parsimonious
model that gives coverage over a wide range of people and expressions.
This is difficult because a morphable model is essentially a linear basis
fitted to nature's nonlinear shape (and/or texture) function. I'll survey
some more sophisticated approaches to the shape function, the new
functionalities they offer, and the technical challenges they present for
computer vision.

Bio of the Speaker: (TBA)

Host: Stan Sclaroff (http://www.cs.bu.edu/~sclaroff)