Faces Everywhere

Why do we see faces everywhere we look: in the Moon, in Rorschach inkblots, in the surface of oil spills?

More than a decade ago, Diana Duyser of Florida, received a religious message through an unlikely medium: a grilled cheese sandwich she had made herself. As she gazed at the brown skillet marks on the surface of the bread, a familiar visage snapped into focus. 8220;I saw a face looking up at me; it was the Virgin Mary staring back,8221; she told reporters in 2004. 8220;I was in total shock.8221;

Such faces made headlines again near the end of 2006, when Mars Express, an orbiter from the European Space Agency, captured the highest-quality three-dimensional images to date of what looks like a face in the Cydonia region of Mars. The photos reignited conspiracy theories that governments on Earth are trying to hide the existence of intelligent life on Mars.

Why do we see faces everywhere we look: in the Moon, in Rorschach inkblots, in the interference patterns on the surface of oil spills? Why are some Lay8217;s chips the spitting image of Fidel Castro, and why was a cinnamon bun with a striking likeness to Mother Teresa kept for years under glass in a Nashville coffee shop, where it was nicknamed the Nun Bun?

Compelling answers are beginning to emerge from biologists and computer scientists who are gaining new insights into how the brain recognises and processes facial data. Long before she had heard of Diana Duyser8217;s grilled-cheese sandwich, Doris Tsao, a neuroscientist at the University of Bremen in Germany, had an inkling that people might process faces differently from other objects. Her suspicion was that a particular area of the brain gives faces priority, like an airline offering first-class passengers expedited boarding.

8220;Some patients have strokes and are then able to recognise everything perfectly well except for faces,8221; Dr. Tsao said. 8220;So we started questioning whether there really might be an area in the brain that is dedicated to face recognition.8221;

Tsao used functional magnetic resonance imaging to record which areas of the brain were activated when macaque monkeys were presented with stimuli including fruits, gadgets, scrambled patterns8212;and faces. She discovered almost immediately that groups of cells in three regions of the brain8217;s temporal lobe seemed to be strongly attuned to faces.

8220;The first day we put the electrode in, it was shocking,8221; Tsao said. 8220;Cell after cell responded to faces but not at all to other objects.8221; Her results were published in October in the journal Science.

Story continues below this ad

Tsao8217;s investigation yielded a surprising related finding: areas of the brain she had identified as face-specific occasionally lighted up in response to objects that bore only a passing resemblance to faces.

8220;Nonface objects may have certain features that are weakly triggering these face cells,8221; she said. 8220;If you go above a certain threshold, the monkeys might think that they8217;re seeing a face.8221; In the same way, she said, objects like cinnamon buns, rocky outcroppings and cloud formations may set off face radar if they bear enough resemblance to actual faces.

Pawan Sinha, a cognitive scientist at the Massachusetts Institute of Technology, has devoted years of research to figuring out just what attributes touch off these face-specific pings. Security software that is being developed for identifying potential terrorists or detecting intruders must be able to reliably recognise faces. In teaching the software to do this, Dr. Sinha and his colleagues have arrived at unexpected insights into why we sometimes see a cinnamon bun as a cinnamon bun, and other times as the earthly incarnation of a sainted nun.

To develop detector software optimised to pick out any human face, even in less-than-ideal surroundings, Dr. Sinha began by putting into his computer hundreds of faces as varied as those in a Benetton advertisement famous for its diversity.

Story continues below this ad

As the computer amassed the information, it was able to discover relationships that were of great significance to almost all faces, but very few nonfaces. 8220;These turn out to be very simple relationships, things like the eyes are always darker than the forehead, and the mouth is darker than the cheeks,8221; Dr. Sinha said. 8220;If you put together about 12 of these relationships, you get a template that you can use to locate a face.8221;

Most people think of the cartoon smiley face, with its discrete eyes, nose and mouth, as the quintessential face template, but Sinha8217;s computer can identify faces even when the pictures are of low resolution.

When he presented human subjects with blurry face images, containing only 12 by 14 pixels8217; worth of visual information, they performed similarly well, recognising 75 percent of the face images accurately. This suggests that like the computer, the human brain processes faces holistically, like coherent landscapes, rather than one feature at a time.

These images are just 8220; dark blobs on a big blob,8221; Sinha said. 8220;So clearly there8217;s not enough diagnostic information in the individual features. Yet something about the overall organisation of the image, the gestalt, is still allowing us to recognise the face.8221;

Story continues below this ad

Once in a while, the computer emits a false alarm. 8220;This is a good analogy for what the human brain might be doing,8221; Dr. Sinha said. 8220;Like the computer, it8217;s trying to determine what the regularities are in all of these faces to create a prototype. But this prototype is not perfect,8221; he said. 8220;Sometimes genuine faces do not match these regularities, and sometimes nonfaces satisfy them.8221;

In other words, if the pattern of light and dark patches on a brindle cow happens to correspond to our conceptions of what a face should look like, we may interpret the coincidence as a visitation from Jesus Christ or Marilyn Monroe.

ELIZABETH SVOBODA