Imagine being able to adjust your face in a photo
to make it look more striking and memorable before uploading it to Facebook or
attaching it to a CV. These are just two suggested applications of an algorithm
developed by researchers based at MIT's
Computer Science and Artificial Intelligence Laboratory (CSAIL).
We're bombarded daily with new faces, which we
evaluate instantaneously and automatically whenever we see them. Not only does
that mean we make snap judgements about people's' personalities, but our brains
decide almost instantly whether or not we are going to remember their faces.
The algorithm is described in a study, Modifying
the Memorability of Face Photographs, which was presented at the ICCV in Sydney earlier this month, as being
able to make subtle changes to faces so that they are more or less likely to
stick in peoples' minds. Sometimes, as the paper points out, the changes are
imperceptible, but they are enough to have an impact on whether the face gets
remembered or forgotten.
The point of the algorithm is not to use it to airbrush photos of yourself
to make yourself more attractive, but to make yourself more distinctive, so
you're less likely to slip out of people's memories once they've met you or
seen your photo. In fact, the whole point of the algorithm is that it manages
to keep factors like your age, attractive and emotional magnitude stable, but
modify only the memorability.To work out the formula the team relied on previous work, which suggested all of us tend to remember and forget the same faces, suggesting that memorable faces could have certain factors in common. The study refers to other work that has found that caricatured faces that accentuate distinctive facial feature also help improve vulnerability, but because the authors needed to focus on "preserving face identity", they say, the algorithm "is likely to change the faces in more subtle ways than simply enlarging distinctive physical traits".
Primarily the algorithm was built based on scores from a visual memory game, although the researchers had to take into account there is a very high false hit rate, as people regularly think they recognise faces they have never seen before. Rather than memorable, these faces count as "familiar", and therefore the false hit rate was subtracted from the hit rate to be able to calculate the true hit rate.
The team tested the algorithm by showing people photos that had been manipulated to be either more memorable or forgettable, and then quizzed them to see which faces they remembered best. They algorithm achieved a success rate of 74 percent.
Memorable faces, the team discovered, tend to be more interesting, but do not necessarily all have one factor common -- such as age or distinctiveness -- in common. "Essentially, our data-driven approach is effectively able to identify the subtle elements of a face that affect its memorability and apply those effects to novel faces," the study says.
According to its abstract: "quantifying and
modifying the 'memorability' of a face lends itself to many useful applications
in computer vision and graphics, such as mnemonic aids for learning, photo
editing applications for social networks and tools for designing memorable
advertisements." It also suggests that it could be used by animators or
makeup artists to draw more attention to certain characters. Most importantly
though, the authors conclude, they have managed to prove that just like
emotion, memorability can modify faces in minute and subtle ways, and yet
change the reactions of others to it completely.
0 comments:
Post a Comment