Recognizing a Face: Identity Invariance Across Different Viewpoints

How is it that we recognize a person’s face from every angle? It remains a mystery how the brain does this and it’s been extremely difficult to achieve with computers. A paper published this month provides new insights into how the brain may achieve face identity invariance across different viewpoints.

The paper is titled “Functional Compartmentalization and Viewpoint Generalization Within the Macaque Face-Processing System” and was published November 5, 2010 in Science.

Brain imaging shows that, in non-human primates, there are six discrete areas of cerebral cortex that respond specifically to the sight of faces. Researchers looked closely at the physiology of four of these regions. They found that the two areas on the temporal lobes towards the back of the brain responded to individual views of faces. The area on the temporal lobes about midway between the front and the back of the brain responded vigorously to mirror image views of faces. Finally, the area on the temporal lobes closest to the front of the brain didn’t respond vigorously to any one view of a face but responded consistently to different views of a particular face.

These results, in addition to data on when signals arrive at each area, suggest a hierarchical processing of visual signals from back to front along the temporal lobes that results in identity invariance across different viewpoints. The mirror image responses are particularly intriguing. How are they contributing to signal processing?