How does the human brain recognize faces

The differences are usually tiny: the shapes of the nose, mouth and eyes, their distances from one another, the height of the cheekbones and forehead. For a long time, neuroscientists have puzzled over how the human brain can still accurately distinguish faces.

This task is as complex as the American psychologist Robert Yin first found out in an experiment in 1969 that all you have to do is turn a portrait upside down and the facial recognition service in the viewer's mind is overwhelmed.

Because this is not the case with other objects such as houses or cars, neuroscientists previously suspected that the mechanism of face recognition in the brain works quite differently, more complexly and apparently detached from the way simpler objects are processed.

But now the theories based on it start to falter. "The reversal effect does not show that faces demand special performance from us, but only that we are particularly skilled in recognizing them, because faces are particularly important to us.

We perceive them in the same way as other objects - only the brain has so much training in it that we cannot get rid of the familiar processing methods, "says neuroscientist Maximilian Riesenhuber from Georgetown University.

With the results of his latest study (Neuron, Vol. 50, p. 159, 2006) the son of the former German Science Minister Heinz Riesenhuber could steer the research of perceptual processes in the brain in a new direction.

Recognize through training

For people, for example, who suffer from facial blindness (prosopagnosia) and who can distinguish between houses and cars, but not their fellow human beings, there would therefore be the previously unimagined chance that they would learn to recognize faces from the ground up through a lot of training.

According to a survey by Thomas Grüter from the Institute for Human Genetics in Münster, two percent of the population suffer from prosopagnosia. And because this is also often associated with autism, Riesenhuber and his team expect further findings: The national health authorities in the USA are funding his follow-up project on autism research.

Riesenhuber's observations reveal that the brain does not make any complicated calculations in order to determine the size of individual parts of the nose or eyes and their relationships to one another. Rather, a group of neurons specializing in face recognition perceive each face as an "overall figure" and process small distinguishing features in tiny deviations in the group-internal activity - similar to what less specialized neurons do when looking at other objects.

Riesenhuber and his colleagues found this out with the help of a magnetic resonance tomograph, which observed the brains of test subjects while they were looking at computer-generated facial images.

Dogs turned upside down

Riesenhuber's team showed the test subjects two different faces at the ends of a series of portraits. Towards the middle of the row, the two faces became more and more intermingled. "If I gradually align the faces with each other, I see how the neuron activity also aligns when I look at it," says Riesenhuber.

And to the same extent the ability of the interviewees to distinguish the two faces decreases: a type of brain activity that also corresponds to the perception of other objects such as cars, "only when looking at cars is on the one hand the number of neurons involved less, "says Riesenhuber.

On the other hand - apparently due to less training - most people did not have a specialized group of neighboring neurons for recognizing cars or houses. In face recognition, on the other hand, a special group of cells, the FFA (fusiform face area), is active on the cerebral cortex behind the ear.

Until now, neuroscientists had assumed that only this FFA area in the brain made it possible to recognize faces. Facial blind people, in whom the FFA neurons are apparently disturbed, were considered to be evidence of the uniqueness of this mechanism. The inability of humans to recognize upside-down portraits seemed to prove that the FFA works very differently from the recognition mechanisms for other objects.

However, an older study from 1986 has been forgotten, says Riesenhuber, in which psychologists Rhea Diamond and Susan Carey from the Massachusetts Institute of Technology looked at judges who had been examining the physique of pedigree animals at dog shows for years: these men and women could not recognize the body qualities of the animals based on pictures of dogs turned upside down.

"That was the first indication that it is not about the objects of contemplation, faces, dogs, cars or houses, but solely about the experience with an object of contemplation," says Riesenhuber. The recognition of faces seems to be able to be trained accordingly. "Then maybe other brain areas than the FFA take on this task."

Riesenhuber's team has now checked the hypothesis with pictures of cars on ordinary people and proven car experts. "It seems to be confirmed," says Riesenhuber. The auto experts would therefore have more difficulty identifying models that have been turned upside down than other people.

"This inversion effect has also recently been shown to butterfly experts. They could no longer assign the animals so well because their butterfly neurons are tuned as sharply as the FFA neurons in ordinary people."

"These results are interesting," says Hanspeter Mallot from the University of Tübingen. "Because they show that face recognition is also a matter of practice." This could also explain, for example, why Europeans find it difficult to distinguish Asian faces and, conversely, Asians find it difficult to distinguish European faces. "You don't have that much practice at it."

Nevertheless, Mallot warns against ignoring the uniqueness of this achievement and its coupling to a separate, special brain area. "Babies, for example, do not yet show any specificity for upright faces and thus no inversion effect." Presumably, the brain area FFA will only be trained to perceive upright faces later.

It remains to be seen whether, in the event of damage to the FFA, other brain regions can also take on the task of the highly specialized group of neurons and those with prosopagnosia can be helped through training.

"This study simply claims that the recognition of faces is not a uniquely complex task," criticizes neuro-computer scientist Christoph von der Malsburg from the University of Bochum. "But we need special functions for this. We can recognize faces, for example, even with different lighting and perspectives, and only because our brain creates face models in a highly complex way - a huge effort that is completely ignored in Riesenhuber's study."

The co-author of the study, Volker Blanz, who teaches computer graphics at the University of Siegen and generated the digital portraits for Riesenhuber, sees a great benefit in the alleged simplicity of human face recognition - but for computer science: "We can learn from the brain that Face recognition must obviously also be technically easier to solve than previously assumed. A simple code, so-called spars coding, should be sufficient to program future computer systems for face recognition. "

How exactly cannot be answered yet. But he and his colleagues wanted to pursue the idea.