Eidos Never Forgets A Face

IN 1988, UNIVERSITY OF SOUTHERN CALIFORNIA neurobiologist Christoph von der Malsburg was pursuing a theory of how the visual cortex in the human brain processes information. When a person searches for a face in a crowd, he reasoned, the brain must strip away meaningless variations, focus on essential features, and confirm their relationships.

To test the theory, he and his colleagues at USC and Bochum University in Germany devised software that reduces images in a similar fashion. The system, called Eidos, worked so well that Germany's Deutsche Bank adopted it to beef up personnel security at its central computing facility in Dusseldorf. Other versions may soon be on sale in the U.S.

Eidos begins by scanning a facial image to find 45 landmarks--including eyes, nose, and lips--and frames them in a grid (photo). It describes those landmarks in terms of mathematical features called "wavelets," which emulate visual cells' responses in the brain. The software then compares the target image with other faces in a database. Deutsche Bank's system, running on a Pentium PC, takes just three seconds to verify the identity of the face, matching it against a stored template. The U.S. Army, which funded the USC research, is evaluating Eidos against competing recognition systems, including one developed at Massachusetts Institute of Technology's Media Laboratory.