How the brain recognises a face

The way in which the mind deciphers faces to tell us apart comes down to just a few hundred nerve cells in our brain's temporal lobes and a code that is "extremely simple", two US scientists announced this week.

The ability of humans and other animals to deftly tell a known face from a sea of strangers has baffled scientists for decades.

Researchers had even hypothesised the existence of individual nerve cells for every face we know, like the "Jennifer Aniston" neurones famously documented in the hippocampus, where other types of memories are laid down. But this turns out not to be true, say Caltech researchers Le Chang and Doris Tsao who think they've finally unpicked the process.

Working with two monkeys, Chang and Tsao used implanted electrodes to record from nerve cells in the lower part of the animals' temporal lobes. In this region are what the researchers dub "face patches", clusters of neurones - or "face cells" - which dramatically alter their patterns of firing activity when an individual looks at a face.

But rather than a one-cell per face arrangement, the studies on the monkeys showed that actually each face cell stores information about one aspect of the face, which might be the distance between the eyes, the colour or texture of the skin, the height of the hairline, and so on.

Different cells respond by different degrees, to the scales or extents of each of these features.

So if that feature is present, regardless of the face that is being presented, then that cell will fire. And it's the ensemble of cells that fire off together when a face is presented that encodes the identity of the face.

Presenting their data this week in the journal Cell, Chang and Tsao proved this was the case by using manipulated images of a face in which different parameters were varied while others were kept the same.

The technique was so successful that they were able to use the pattern of brain activity they recorded to reconstruct an image of the face that the animal had seen. Shown to a human, a person could pick out the original image the animal had looked at.

The researchers suggest that their findings have implications for developing better artificial intelligence systems to machine recognition of faces and other objects.