Researchers develop privacy filter to block photo facial recognition
A team of researchers at the University of Toronto Engineering school have used adversarial training to create an algorithm which dynamically disrupts facial recognition systems, Tech Xplore reports.
The team was led by Professor Parham Aarabi and graduate student Avishek Bose, and designed a pair of neural networks; one to identify faces and the other to change images in order to disrupt it.
“Personal privacy is a real issue as facial recognition becomes better and better,” says Aarabi. “This is one way in which beneficial anti-facial-recognition systems can combat that ability.”
The resulting algorithm changes specific pixels within the image to make alterations nearly imperceptible to human eyes, with what Tech Xplore calls an Instagram-like filter. Tested against the 300-W dataset, it reduced the success rate of face detection from nearly 100 percent to 0.5 percent. The technology disrupts all automatic extraction processes for facial attributes, according to Tech Xplore.
“The disruptive AI can ‘attack’ what the neural net for the face detection is looking for,” says Bose. “If the detection AI is looking for the corner of the eyes, for example, it adjusts the corner of the eyes so they’re less noticeable. It creates very subtle disturbances in the photo, but to the detector they’re significant enough to fool the system.”
The team will present its findings at the 2018 IEEE International Workshop on Multimedia Signal Processing, and hopes to make the filter available through an app or a website.
Google researchers have previously created adversarial images that can defeat AI image recognition systems, while Facebook is currently locked in a legal battle over the privacy implications of its faceprints photo tagging system.