Researchers at Carnegie Mellon University have managed to develop a set of glasses capable of fooling commercial-grade facial recognition software into identifying you as somebody else entirely. They cost $0.22 (18p) to make, and worked with unnervingly high success rates in tests.
Facial recognition software relies on deep neural networks, which learns how to tell individuals apart by focusing on specific features, such as nose shape and eyebrows.
Not only can the glasses cover these up, but as they were printed out on photo paper, they can be marked with specific printed patterns that trick the software into seeing the facial features of a completely different person. In tests, the researchers managed to disguise non-famous people as famous people they look nothing like, such as John Malkovich, Milla Jovovich and Colin Powell.
They pitted their creation against Face++, the system used by Alibaba for secure payments, by digitally editing the glasses onto a picture, and fooled it 100% of the time, though success rates could be lower in ‘real-world’ scenarios. [Quartz]