August 22, 2016 -
In a presentation delivered at the Usenix security conference earlier this month, security and computer vision researchers from the University of North Carolina said that would-be criminals could exploit Facebook images to dupe certain facial recognition technology, according to a report in Yahoo Tech.
In its research, the UNC researchers collected photos of 20 volunteers from various websites including Facebook, LinkedIn, and others.
Using these images, they generated 3D models of each person’s face, added facial animations, and edited their eyes so that they were directly facing the camera.
For those subjects whose images did not display their full face, the researchers recreated the missing components of the face as well as embellished with shadows and textures.
In some cases it was incredibly difficult to find images of the subjects online. Despite this, the researchers were able to use a few low-resolution images to generate a model accurate enough to trick some facial recognition systems.
Overall, the UNC team successfully tricked four of the five systems with success rates from 55 percent to 85 percent.
“We could leverage online pictures of the [participants], which I think is kind of terrifying,” said True Price, a computer vision researcher at UNC, in an interview with Wired. “You can’t always control your online presence or your online image.”
All five of the systems tested — KeyLemon, Mobius, TrueKey, BioID, and 1D — are currentlyavailable on the Google Play Store and the iTunes Store.