Neuro-Science Thought Experiment on Facial Recognition

Friday, October 10, 2008

This isn't really a post as much as me just writing down a future reminder-note of what I think would be an interesting research project involving facial recognition in the brain. This note will remind me to Google it later and see if it exists already.

My thought is that if we were shown a picture of a total stranger who is standing in an area very familiar to us, then we would more easily recognize him or her later in public than if we were shown a picture of him or her in an unfamiliar area.

The idea behind this is similar to size reference in photographs: e.g. if we're shown a picture of a rock, we have no idea how big it actually is until we see something familiar next to it, like a hand (or even better: our own hand).

Now imagine you see a picture of a stranger, or even a celebrity. When you see this person in real life, you almost immediately notice how their face is different than you expected. This is because the mapping from 2 dimensions into 3 is not perfect, and a lack of reference in the photos you saw, I'm guessing, makes it less perfect.

If you instead see a picture of this person in your front lawn or in your room, then your extreme familiarity with the reference points should correct some of the subtle flaws in bone structure, even lighting. In other words, you subconsciously know how things "should look" in your familiar environments and inversely how things would look outside of the environment.

So it would be interesting to get 3 groups of people; the first group is shown pictures of a stranger in unfamiliar areas, the 2nd familiar areas, and the 3rd a mix. Testing recognition would be the hard part. I would choose one of these:

1. Have a line-up of strangers that all look very similar to each other and see who correctly guesses the right one.

2. Have the stranger walk into the testing facility just as the subjects leave and see who noticed him or her.

3. Show altered photographs of the stranger and see who most accurately "senses" the differences.

The applications for computerized facial recognition and 3-d mapping from a 2-d image are also interesting. Theoretically, you could thoroughly familiarize a computer with a space (say an airport) and use that knowledge to more accurately recognize faces of criminals walking about. Or beyond that, allow a computer to more quickly generate an understanding of new objects in a space by using pre-processed information to reduce the processing of real-time recognition.