Path: utzoo!utgpu!watmath!clyde!att!osu-cis!killer!ames!ucsd!rutgers!iuvax!ndcheg!uceng!dmocsny
From: dmocsny@uceng.UC.EDU (daniel mocsny)
Newsgroups: comp.ai
Subject: Generating facial caricatures
Keywords: recognizing faces, exaggeration
Message-ID: <564@uceng.UC.EDU>
Date: 8 Jan 89 22:56:12 GMT
Organization: Univ. of Cincinnati, College of Engg.
Lines: 28

Several years ago I read about some image processing techniques for
generating facial caricatures from facial photos. The researchers
responsible expressed the geometry of a face with a few tens of
variables. They then measured these variables for several hundred
photographs of different faces, from which they developed composite
male and female faces. Then for any given face, they computed how each
of its variables differed from the corresponding value for the
composite face. They multiplied each difference by some constant, and
used the exaggerated valued to generate a caricature.

An interesting result of the work occurred when they generated
caricatures of famous faces. They found that test subjects could
recognize the caricature faster than they could the original. Thus
they concluded that our brain remembers a face by a process similar
to computing how it differs from the average face. This may also
be at the root of a common observation between two dissimilar
ethnic groups that meet for the first time: each group thinks the
members of the other group all look very similar. Perhaps some time
must pass before the members of one group can generate a composite
face for the other group.

Are any comp.ai readers familiar with this work? If so, I would
appreciate the original reference.

Cheers,

Dan Mocsny
dmocsny@uceng.uc.edu