A surprising number of people, from social media platforms to police surveillance systems, are very interested in what you look like. That means companies are able to learn a lot about you — your appearance, age, ethnicity, and more — whenever your face pops up. While this is generally for the sake of targeted ads, it can also put people at risk of privacy violations or identity theft.
To help people hold onto (whatever remains of) their privacy, tools have emerged intended to trip up facial recognition AI. Real-world products like “Face Off Hats” (that’s a physical hat with a trippy pattern) and 3D-printed masks fool face-scanning software by presenting optical illusions to the cameras we encounter during our everyday lives.
But those tactics are to throw off the cameras. What happens if a photo of you somehow makes it online?
Soon, there may be a filter to keep AI from spotting your face those photos that slip past.
Engineers from the University of Toronto have built a filter that slightly alters photos of people’s faces to keep facial recognition software from realizing what its looking at. The AI-driven filter looks for specific facial features and changes certain pixels. People can barely see a difference, but any AI scanning the image can’t even tell that it’s looking at a face.
(Credit: Avishek Bose)
To build its filter, the team pitted two neural networks against each other. The first AI system was tasked with identifying facial features from a set of several hundred photos, and the second algorithm was responsible for altering the photos to the point that they no longer looked like faces to the first...MORE
