Researchers at Carnegie Mellon University have developed a new method of fooling facial recognition software that requires nothing more than a pair of glasses, reports New Scientist.
To be fair, they are really weird glasses, with large, multi-colored frames that don’t adhere to much of a coherent aesthetic. It’s primarily the patterns on the frames that fool the facial recognition software, which in this case was the commercial available Face++ platform, which the researchers call “state-of-the-art” and has attained an accuracy rate of 97.3 percent on a standard performance challenge. Knowing the specific biometric signals that would be highlighted by the facial recognition software, the researchers designed the patterns to mimic the signals of particular celebrity actors – Aaron Eckhart, Brad Pitt, Clive Owen, Drew Barrymore, and Milla Jovovich. In other tests, they designed glasses that the software would erroneously read as races and genders different from the subject’s own.
They were largely successful. While one researcher was only read as intended target Clive Owen 16 percent of the time, others successfully swapped ethnicity and even gender with 90 percent accuracy.
It’s a more sophisticated means of thwarting facial recognition technology than the PrivacyVisor announced by Japanese researchers last year, since while that system sought merely to evade successful identification, this one actually allows for impersonation. But it’s also limited in its application: A fully automated security system depending solely on facial recognition could be fooled, but a human border guard isn’t going to mistake a man wearing goofy glasses for Milla Jovovich. As such, the system points to the need for multi-factor authentication, if not a bona fide human element, in sensitive security systems.
November 3, 2016 – by Alex Perala