Re:TFA (Score 3, Interesting)

Some years ago (I've long lost the reference) a PhD student in Rhode Island digitized human faces with 256 points and then projected these points through the same points on an androngenous composite of hundreds of faces next to the sample; an "average" face. At some point in a plane beyond the reference face, the points at the ends of the projectors were then re-plotted and joined to form a caracature of the amplified differences between the sample face and the sexless "norm". She showed that her subjects were much more likely to recognize the characature than they were the actual face, and postulated that facial recognition therefore depended on a similar process. One of her samples was Ronald Reagan, but that's the only basis in time I can remember.