CSAIL Event Calendar: Previous Series

Biometrics and Isometric Invariant Measures

Speaker: Ron Kimmel , Technion, Israel
Date: August 7 2006
Time: 3:00PM to 4:00PM
Location: 32-D507
Host: Polina Golland, MIT/CSAIL

Contact: Polina Golland, x38005, polina@csail.mit.edu
Relevant URL:

We will discuss the problem of matching metric spaces by embedding
their intrinsic geometric structures in various domains. As for
applications, we start with lip reading by flat embedding (joint with
M. Aharon), and then continue to the 3DFace project (joint with the
Bronstein brothers), and consider expression invariant face
recognition.

We review some theoretical problems and recent advances like the
generalized multidimensional scaling (GMDS) that we use to embed the
geometry of one surface into another.

As a side story, I will comment on the role of biometrics in biblical
stories and fairy-tales.

=================

Most of us are familiar with the story of the prince who declares that
he will marry the girl whose foot fits into the glass slipper she lost
at his palace.

The prince finally locates Cinderella by matching the slipper, and
they live happily ever after. The first version of the story
originated in China around the days of the Tang Dynasty. There,
Yeh-Shen (Cinderella) had the smallest feet in the kingdom, an
indication for beauty in the Chinese culture. In the German version
of the story, one of the stepsisters fits into the slipper by cutting
off her toe, creating a fake biometric signature. But Cinderella's
stepsister was not the first, biometric identification and biometric
frauds date back to biblical times.

In Genesis story, Jacob stole his father's, Isaac, blessing as the
privilege of the eldest, by pretending to be his firstborn brother
Esau. By ``hand scan'' Isaac wrongly verified his son's Esau's
identity, who was actually smooth-skinned Jacob with kidskin wrapped
around his hands.

Face recognition, an important biometric tool, also plays a major role
in fairy-tales. For example, facial features are examined one by one
before the unavoidable conclusion is made by Little Red Riding Hood,
its the wolf she was talking to, rather than her grandmother. We leave
fairy-tales for a while, and try to introduce some axiomatic arguments
that will help us decipher the face recognition enigma. That is, we
identify the measurable invariants we could use in the context of face
recognition.

When we reach adulthood, the global or integral measures of our body,
like height and weight, change relatively slowly over time. As we move
our limbs and bend them at their joints, like wave our hands or make
facial expressions, the skin surface stretches just slightly. We could
therefore exploit the geometry of the outer envelope of our body, that
is, the skin surface, as an invariant for recognition purposes.

What are the facial measures we could use in order to tell one person
from another? Our stating point is the fact that our skin surface
can, in most cases, be modeled as an isometry. We validate this
property empirically on the most complicated outer part of our body:
the face. After proven valid for the face, our isometry model would
obviously hold for other ``simpler'' objects, like hands, legs, and
feet. It would eventually allow us to help our prince build an
automatic procedure to locate Cinderella.

See other events that are part of Biomedical Imaging and Analysis 2006/2007

See other events happening in August 2006


About Us Research News Resources Directory