We all know somebody with one of those faces. You know, the friend who always gets mistaken for someone else. They say, “I know I remember you from somewhere!” But they don’t. Turns out there’s a science to this sort of thing — and it could make your face more memorable.
A team of researchers at MIT’s Computer Science and Artificial Intelligence Laboratory recently developed an algorithm that can make photos of faces easier to recall. The algorithm makes slight tweaks to the size, shape, and appearance of the face; then, almost like magic, people remember the adjusted photo of the person more often. This is not pseudoscience or sorcery. It’s maths.
The experiment used to test this new algorithm is remarkably simple, though incredibly exhaustive. First, the researchers built a database of over 2,000 random faces. They then assigned each of the photos a memorability score by showing Amazon Mechanical Turk workers hundreds of photos and asking them to flag the faces they remembered. This led to each photo receiving a “memorability score” that a computer then used to identify specific features in the photos. It wasn’t necessarily whiter teeth or blue eyes, however, but tiny mathematical differences in the shape and appearance of the face, as well as adjustments to attributes like age and attractiveness, that affected a face’s memorability.
The algorithm basically takes over from there. The researchers selected 500 photos and fed them into the program which then created thousands of slightly modified copies. The more memorable versions were selected over the less memorable, feeding the algorithm more information about what makes for more memorable faces. That process was repeated again and again until the software had determined it had reached an optimum level of memorability.
Turns out the software works. When the enhanced photos were sent back to the Mechanical Turk workers, they were indeed more memorable about 75 per cent of the time. But, again, there was no silver bullet to making a face easier to recall. It requires a process and lots of tiny changes.
It’s unclear where the algorithm will go from here. OK, fine: Facebook will probably buy it. [Gizmag]