Monday, July 31, 2006

Crossing the Uncanny Valley

Back in 1970, a Japanese scientist named Masahiro Mori found that people would accept robots that looked like robots, or that looked like humans. But somewhere in between, when robots began to look almost, but not quite human, most people felt a sense of unease, even revulsion. Mori called this the “uncanny valley” effect.

It’s so well known among digital animators that they routinely add imperfections to the faces they create for movies and computer games, in order to avoid upsetting the audience.

Steve Perlman has a better idea. Why not make computer-generated faces look exactly like the real thing, down to the last wrinkle? Perlman says he can do it with his new technology, called Contour. Based on the videos he’s showing off here at SIGGRAPH, it looks like Perlman has nailed it.

Contour works sort of like the motion capture systems used to let characters in games move like real people. But the system goes further. By coating an actor’s face in phosphorescent makeup, Contour can capture many thousands of glowing dots, corresponding to the coordinates of the face. By scanning the actor as he talks, laughs, shouts and so on, it’s possible to map the fine details of the face with extraordinary precision. Then computer animators can use that data to create an all-digital rendering of the face. The result looks exactly—exactly—human. For those of us who’ve spent years playing video games that haven’t gotten the faces quite right, the effect is remarkable.

Perlman said he’s working with Hollywood film producers and computer game designers to create the first Contour-based entertainment products. Sometime in the next year or so, digital artists will start crossing that uncanny valley, and I doubt they’ll ever look back.
(By Hiawatha Bray, Globe staff)

Posted by at 05:49 PM

Print | E-mail to a friend | Permalink | Subscribe via rss