Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Thoughts on face-recognition

2 views
Skip to first unread message

Jorn Barger

unread,
Sep 2, 1995, 3:00:00 AM9/2/95
to

Face recognition is an interesting problem, partly because its brain
circuitry must have been evolving for hundreds of millions of years...

I've proposed an AI-complementarity principle that states that the
knowledgebases used for *comprehending* phenomena should usually also
be capable of *generating* simulations of these phenomena.

And I've cited a corollary to this, based on the fact that the
*simulations* can be simplified to any level, that the comprehension-
module should also be possible-to-build incrementally from such a
simplified level...

In the realm of face-recognition, natural selection must have placed
great importance on recognizing faces from any angle, and at great
distances. And the shape of the face would have co-evolved to make
this easier, along with the use of facial muscles to express
easily-identifiable emotions. (My theory of the evolution of eyebrows
is that their function is primarily for signalling emotion.)

So the neurons involved in face-recognition should have some
representation-scheme that abstracts out some basic curves, able to
turn them in any direction, and to identify them even when only a very
small portion of the retina is affected...

They must also be able to take advantage of shadows-- one curve mapped
across another! (It's an interesting experiment to turn the contrast
way up on a tv, and the brightness way down, so that all you see are
patches of light and shadow, and notice how little recognizability is
lost. This trick can also distinguish skillful use of makeup from
unskillful...)

Surely the representation scheme for these curves must take account,
in humans, of the musculature of the face, as well?

So the deformations of a curve might be represented in terms of the
sorts of muscular tension that could produce them-- even, possibly,
when the deformations are produced not by muscle but bone.

So someone whose bonestructure gave them a pinched face might be
perceived as having a pinched *personality*, because our memory
scheme remembers their facial curves *as if* produced by pinched
musculature...


j
jo...@mcs.com

0 new messages