Want To Understand How Facial Recognition Works? You'll Need A Greeble

Illustration for article titled Want To Understand How Facial Recognition Works? You'll Need A Greeble

Facial recognition is an important and interesting subject. It's also a difficult subject to study. To help break down how facial recognition works, you need a greeble. Or better yet, you need whole families of greebles.


One of the oddest discoveries in the science of facial recognition is that people recognize faces. That sounds redundant, but it's not. When you see your best friend's face, you recognize it. You're not as likely recognize your best friend's nose in the mix with a lot of noses. You're also not likely to recognize your best friend's nose on a face that's been slightly altered - the eyes spaced farther apart or the mouth slightly thinned. You recognize your friend's face because of their particular configuration of features. Each feature on its own is strange to you.

To compare the mechanisms of facial recognition to the mechanisms of other kinds of recognition, researchers need to be able to compare faces that can be distinguished by certain features with other objects that can be distinguished by certain features. For some time, researchers used animal photos. They might try to see how subjects in experiments can distinguish between two birds that have share some features - they're both wading birds - but have some features that are wildly different - one has a particular crest on its head and the other doesn't. They also might ask people to distinguish between different members of the same species.

Illustration for article titled Want To Understand How Facial Recognition Works? You'll Need A Greeble

Studies like these ran into problems because people have different levels of experience with animals, and when people do have experience, that experience can screw up the test. Researchers needed an object that was entirely novel but had a number of features that could be varied during the experiment. They came up with greebles.

These greebles come in two "genders." "Plok" has drooping features. "Glip" has upturned features. Greebles come in five "families," each with a different body shape. "Galli" as we can see, has a shape that looks a bit like a teapot, while "osmit," right next to it, looks more like a flower pot.

The protruding horns are the features that make each greeble an individual, the way noses and eye shapes make us individual. The horns on each shape's head are called "boges," the upper horn on its face is a "quiff" and the lower one is a "dunth." A horn can be oval or triangular, it can be a long twist or a gently rounded mound. Researchers manipulate the horn shapes to make greebles that experiment subjects haven't seen before, but that still belong to a gender and family.


If this sounds complicated, it's supposed to be. Greebles are meant to test our recognition. Subjects are challenged to recognize a greeble's gender and family, both when it's one of the greebles from the grid or a new greeble. It's a way to see if experiment participants recognize shapes the way we recognize faces - in pieces or as a whole. And if they don't recognize greebles on sight at first, how long before they, like the psychologists who invented the greebles, become greeble experts, sorting them into family and gender no matter what new boges they wear?

You can test yourself quickly by looking at the top greeble again. Can you, on sight, tell its gender and family? Would you recognize it if you were to see it again?


Images:Scott Yu, I. Gauthier, I., M.J. Tarr

[Sources: Becoming a "Greeble" Expert.]



I have face blindness, so recognizing faces for me is a lot like recognizing greebles - it can be done, but it takes a lot of work or a lot of familiarity to commit what certain features look like to memory. I'm also less likely to recognize a person's face if seen at an unusual angle, or in unusual lighting - which is probably what most of us would experience trying to recognize greebles.

People ask me what it's like, and I like to answer that it's totally normal, it's just that everyone else is superhuman. Most brains are equipped with certain special neurons that help them to specifically recognize faces, whereas I see faces for the collection of shapes and colors that they really are. If I put upwards of 100 arbitrary, but similar, collections of shapes and colors in front of you, then asked you to recognize them from different angles and in different contexts, you'd have trouble, too.