Facial Recognition

HERE’S LOOKING AT YOU KID, TAKE 2

We are born with a special ability to recognise faces. It is such a powerful ability, that we see faces everywhere.

 

Every time I sit at my office desk, I am looking at the grey slate tiles on the floor with their white random streaks throughout, and I see on one of them a face looking back at me. No one I know luckily. There are patterns all over the tiles, but this one spot on that one tile draws my attention like a magnet.

When you are born, you arrive in a world that cannot be anticipated.  So we are born with a body and a brain with all the flexible potential to cope with whatever we may encounter. It takes time, especially for a human baby born so underdeveloped, to learn what will be safe, so years of trial and error while learning, all the time nurtured and protected by someone older and more experienced.

Most of what we have evolved to be, has occurred over millions of years of experience with genes along with its corporeal host, surviving or not, depending on their fitness for the times. For the last few million of those years, the environment has been pretty much the same, so the genes have had time to settle to a suitable set.

That is up until the last few millennia, when the pace of change of the societies we are born into has been faster than evolution. Faster than the rate of mutation, and weeding out of inefficient ones. The agricultural revolution, the rise of religion, the renaissance, the industrial revolution, the technology revolution. That means the genes best suited for survival, or success, in today’s world, are still experimenting. In fact during times of dramatic environmental change, where the old genes start to have less effectiveness, genetic variance increases. Trying to find out what else might work best.

But that is not the whole story.

Some things are so constant in the world we will emerge into, that it is beneficial for genes dealing with those factors be rigidly established. That is not farfetched. Genes for certain metabolic processes are so crucial and so intolerant of alteration, that those so called “ancient” genes are set in stone in our chromosomes. They’re called fixed alleles. Those that genes that code for proteins that use sugar for energy for instance, the fundamental means of living. Mess around with that and you hardly survive conception.

   

One thing that all higher animals have is a face. In an earlier blog I talked about the evolution of intelligence and how bodies that became bilateral became the norm. (See Blackjay’s “A Brief History of Intelligence”.) And those animals with faces are bilateral animals. Even the weird ones like octopi or insects or snails. Yes I know, an octopus is not bilateral, but hey, they defy any stereotype. So faces are bilateral, even the octopus’s in part. Most faces have two eyes, two nostrils, two ears, one central mouth. That seems for the most part to be the components that constitute a face. Harvey Ball, creator of the Smiley Face knew that.

When Humphry Bogart was looking at Ingrid Bergman, (see Blackjay’s “Here’s Looking At You Kid”) that is what he was looking at. Well probably some other things too, but the pull to see a face is so strong, that that is what probably happened first in his brain. The evidence is it happened before he even knew it, before he saw the rest of whatever his visual cortex had derived from the input from his eyes.

Baby’s can recognise and react to a smile. By about 2 months they smile back. By then they have managed to focus and coordinate their vision functions. At 4 months the baby can distinguish its mother’s face from others. At 4 months the baby can do little else, so it must be vital to do that. The reason this face recognition trait comes into play so early is because it is built in. There is no learning curve, no training by experience. There is a built-in face detector, saves a lot of time since every single human born needs that skill. By contrast, there are no special detectors for say cars or fruit. There is a reaction to sound and to its direction, and to discomfort or pain, but they are reflexes, no learning involved, and the well-known grasping, rooting and Moro reflexes, also innate for early survival. But facial recognition is a higher level cognitive function way above simple reflexes.

It is a conundrum as to why face recognition should be there with those other reflexes which are clearly for survival. The benefits of rooting for milk, flailing around when falling, and grasping to hang on, and such, are obvious. By how would being able to recognise a face, any face, be of equal benefit to survival?

Humphry Bogart saw Ingrid’s face within 40 milliseconds. That is faster than all of those reflexes, faster than even the 100 milliseconds of spinal reflexes which only have to go in and out of the spinal cord.

How does this all happen? Many other primates have it, so that may be a clue. Apes like chimpanzees and monkeys like macaques for instance. Clearly whatever the advantage, it must be an ancient one. Macaques separated from humans 30 million years ago and Chimps more than 4 million.

If we have a special visual ability to recognise faces, then there is presumably a dedicated part of the brain that does that. And indeed it turns out that there is. In some ways, in a conceptual sense, this is not unlike the groundbreaking research by Lettvin, Maturana, McCulloch and Pitts who discovered that a frog’s eye is specialised to detect bugs, the one thing in its environment that is crucial for its nutrition and survival. Silly to have a young tadpole have to learn through trial and error what an edible insect looks like.

 

To understand just how special this is, let’s start with a recap of how the system works in general.

Illustrating just how different the human visual system works compared with every current artificial attempt to duplicate the performance, and hopefully removing some of the hubris behind researchers who think they are “close” to human-like intelligence, the ways our visual processing works are bizarre.

Basically it breaks the visual scene down to fragments then rebuilds them in a way unlike the photograph image we think we see. And the broken-down pieces are not the pixels typically associated with the structure of an image. And the reconstruction generates additional information like location, motion, emotion, recognition, purpose. Much more than what is in a photographic image. This is clearly understood when you look at a confronting image for example. Your processing of a photo of a distressed baby or a tortured individual, or a sumptuous meal, or a sensuous body, results in much more than just the image, yet it is all generated from the visual system. There is no screaming or crying or aromas or stimulation in the photo you are looking at. Your visual system draws all of that extra stuff out just from the visual input.

Visual information enters the eye and undergoes some processing immediately. The retina is part of the brain. All nerves cells can be classified loosely as peripheral (afferent like sensory heading into the brain and efferent like motor heading out) or central (those making up the grey matter of the brain). The retina is composed of central neurons and converts the image falling on it into the strangest first breakdown. In receptive fields, which are circular areas that for simplicity compare if light in their centres is different from light in the outer part. So called on and off centres. And that is basically all that gets through to the next part of the visual processing at the primary visual cortex. Yet from that basic breakup, this next stage is able to reconstitute all the lines and edges in the original image.

     

One might wonder why go to the trouble of this breakdown and rebuild process. The answer is that there a 120 million optical detectors in the retina, and for all of them to relay individually to the next part of the brain, all the way from the eye at the front to the visual cortex at the back, would require such a thick cable (optical nerve) that the flexibility of the eye movements would be hindered.  The optic nerve only needs a tenth of the number after the receptive field processing. So why not put the visual cortex at the front then and make it a short path? Same reason that the motor cortex for the left side of the body is on the right side of the brain and vice versa. An injury to one side of the brain would prevent a motion to react if that same side of the body was then unable to move because of the brain injury. More chance of survival if the visual part of the brain is away from where an eye injury might occur.

From the retina the part that will eventually become conscious goes to the Primary Visual Cortex at the rear of your head, where further processing allocates visual information such as orientation of lines or colours into structured columns. If you looked at all of that and mapped it out, you would already struggle to reconstruct the original image. But your brain does it easily. Then onto the Secondary Visual Cortex where more complex versions of the same features are assimilated. Instead of neurons responding to just one line of orientation for instance, now there are neurons that respond to specific combinations of orientation. For instance the three orientations of the sides of a triangle, or the combination of a circle. Next these basic features start to be associated with actual objects, a circle and lines such as an image of a clock. Then further processing to determine movement and location. This is where the dynamic aspects of the image you are watching are extracted. You know that a hat is located on a person’s head and will move along with the head, while from a picture of a person throwing a ball, you know they are not tied together in space or time. And still higher feature extraction, not the basic image parts like edges or colour as we are familiar with in artificial vision systems, but cognitive features, such as are we late for an appointment when we look at the time on the clock, or damn I missed Neighbours again.

Clocks have been around for a thousand years or so, and are not vital for survival. So there is nothing special about them, and no value in having all of that “clock” detection prewired in. We can learn it on the fly as we grow up.

But faces are different. Two absolutely critical aspects associated with faces contribute to survival. And therefore to the survival of the genes that give this ability. Firstly it allows the person to recognise friend from foe. That alone is sufficient for the benefit over time, the fitness of those genes, to earn their keep. But to add value, being able to detect whether a face is angry or happy helps evaluate the risk of both friends and foes. A third benefit from facial recognition is even more valuable and which has come to the fore in humans. It allows communication. We literally say, read someone’s face. Or as Nixon said, “Read my lips.” Maybe this is a vestigial pathway to language in humans?

All of the usual processing (clocks and what have you) takes place through a couple of established routes called the Dorsal and Ventral Pathways. We have know great detail about the first step, the retina, and the second step, the Primary Visual Cortex, for a long time, yet we are still discovering new processes and pathways. Some areas like V7, we have no idea about. Colour comes out separately in V8 and motion and control at V5, but those areas are poorly understood still.

  

The face recogniser bypasses and avoids all of this and branches off to a region called the Fusiform Gyrus (because it is spindle shaped). And consistent with the evolutionary determination of the criticalness of this visual ability, it is tucked away deep and underneath the brain safe from peripheral injury.

 

Next time someone says they are attracted to something other than the face, it might be true, but it is the face that they saw before anything else.