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

Date: Monday, 11/12/01
Time: 3:30-4:40
Place: NSH 1507

Speaker: Terence Sim
Title: Combining Models and Exemplars for Face Recognition: An
Illuminating Example

Abstract:

After three decades of research in computer face recognition, there are
a number of commercial products that claim to do the job.  These
products actually work, but under controlled conditions.  Specifically,
the imaging conditions of the training image stored in the database and
those in the task image to be recognized have to be similar.  If
lighting or pose changes, for example, then these products fail.
Research aimed at improving the robustness of face  recognition systems
typically use either models alone or exemplars alone to compensate for
changing illumination or head pose.  This has met with limited success.
In this talk, we propose a combined model- and exemplar-based approach
for face recognition.   Our idea is to use models to synthesize many
more exemplars, which are then used in the learning stage of a face
recognition system.  To demonstrate this, we develop a statistical
shape-from-shading model to recover face shape from a single image, and
to synthesize the same face under new illumination. We then use this to
build a simple and fast classifier that was not possible before because
of a lack of training data.