If there is one things humans are very good is face recognition. We can recognise a face under different lightning conditions, we recognise it even if it grew a beard, if the haircut is different, even if it grew older!
Other animals are also able to recognise our face, for sure dogs and primates and possibly several more species. And, I presume, many species are able to recognise a face of an animal of their species, much better than us, of course, who have problems in distinguishing, say a duckling from another whilst mommy duck seems to be pretty good in recognising her offspring.
Scientists (as well as philosophers) have wondered what is the magic through which we can recognise a face and most assumed it was a quite complex endeavour.
Computer scientists have found a few ways to identify a face (biometrics) by using a number of points in a face and measuring their relative distance and area characteristics.
It turns out that mother Nature took a very similar approach that led, at least in the case of primates, to a very effective solution. A set of 205 neurones that fires based on the distance between points in the face. It is like a face is projected onto a multidimensional space with individual neurones activated in a way that is proportional to the projection.
This discovery was done at Caltech by a group of researchers that have published their results on Cell. The discovery is substantiated by a number of experiments in which the researchers by looking at the firing neurones were able to create the image of the face (see the picture) and this image is amazingly accurate with the original that caused the neurones to fire.
I am really amazed by this result, possibly because it was a strong curiosity I always had since I was a kid, to understand how our brain can recognise objects.
The discovery and experiments are related to primates but there is no reason to think that our brain approach is any different (see the clip). Also there is no reason to believe that face recognition is different from the recognition of any other “shape” or object.
It would be nice to identify the neurones in our brain, and see if we are using more or less neurones than a primate. However, the identification involves invasive procedures and it is not easy to find volunteers…
The researchers feel that this discovery can be applied to the booming area of artificial intelligence, providing a way for new learning algorithms and new ways of encoding object characteristics in deep networks.