Algorithm makes your face shots more memorable

By Ben Coxworth

December 17, 2013

Software developed at MIT subtly alters photos to make faces either more (right) or less (left) memorable (Image: MIT CSAIL)

Lots of people wish that they were more attractive, but have you ever wanted to just look more ... memorable? Just a few tweaks here and there, to help keep your face from being forgotten? Well, software created by researchers at MIT’s Computer Science and Artificial Intelligence Laboratory can now make that happen – to photos of your face, that is.

It would be easy enough for the program to make your face more memorable simply by giving it a third eye or coloring it green, but fortunately it doesn't do that sort of thing. In fact, it was designed specifically to alter the subject's appearance only on a very subtle level.

The researchers started with a database of 2,000 facial photos, each of which had already been given a "memorability score" by human volunteers who viewed and rated the pictures. Those images were then analyzed using the software, which was tasked with identifying traits that the higher-scoring faces had in common with one another. It also noted the recurring traits among the lowest-scoring images.

The team then selected 500 faces, and got the software to use its new-found knowledge to create both more- and less-memorable versions of each one. It was programmed to avoid making any changes that would alter the subjects' basic features (such as age and gender), or make them look less attractive.

In each case, the program started by making thousands of copies of the photo in question, each one incorporating slightly different alterations. It then analyzed those copies, and selected the one that best met the criteria. That photo was then copied a number of times itself, again with each copy incorporating different modifications. The best photo of that bunch was once again selected, with the process being repeated until the software determined that it had reached the optimum result.

When the final more- and less-memorable versions of each of the faces were shown to human test subjects who rated them on their memorability, those ratings matched up with software's goals for about 75 percent of the photos.

Suggested uses for the technology include the ability to enhance photos used on social media sites and in job applications, or to lessen the noticeability of the faces of background extras in movies or TV shows.

Source: MIT