One of Pawan Sinha’s projects involves electronically blurring images of celebrities’ faces, then showing them to student volunteers.

The students stumble over some of the blurred countenances. Dustin Hoffman is tough, as is Tom Hanks. But, notes Sinha, “Eddie Murphy is easy; Princess Di is one of the easiest.”

If it sounds like a party game, this and related projects are aimed at shedding light on one of the most fundamental issues in neuroscience: what we’re recognizing when we recognize a face.

Scientists have debated the topic for years. One popular theory is that “fine featural details” — the creases running from nose to mouth, for example, or the exact shape of the eyes — are key. But Sinha’s subjects can often identify celebrities whose blurred faces lack any hint of such details.

“Recognition performance is maintained even at extreme levels of degradation,” notes Sinha, an assistant professor of brain and cognitive sciences.

If not fine details, what do we latch onto?
Asst. Prof. Pawan Sinha is shown with two of the caricatures that figure in his exploration of how we recognize faces.
(Photo: Ed Quinn)

The question isn’t just of academic interest. Computer face-recognition systems are notoriously bad at the types of tasks Sinha’s volunteers perform with relative ease, so progress on understanding how our brains analyze faces could lead to much more versatile artificial systems.

The scientist has already found some of the recognition factors we rely on: the ratio between the space separating the eyes and the width of the head, for example. To further probe the issue, he’s asking 50 caricaturists from publications around the U.S. to do drawings of 100 faces each.

Caricaturists, explains Sinha — himself an artist — have a knack for picking out key qualities in a face. He will supply them with the images of 100 faces, and will analyze the resulting drawings, focusing especially on the facial attributes the caricaturists choose to exaggerate.

Such analyses, applied to the whole face, should offer insights into the brain’s recognition systems. Say there’s someone whose nose and chin are both 10 percent bigger than the norm. If the caricaturists exaggerate the chin — a la standard caricatures of Jay Leno — and draw the nose as it really is, “this would tell us that the distance from the lips to the tip of the chin is perceptually of greater importance than the nose’s length,” explains Sinha.