Obscuring your face does not hide you from facial recognition systems, researchers have found.
A group from the Max-Planck Institute found that blurred images were still individually identifiable with just a few non-obscured images to train from. With the proliferation of images on social networks, it is possible that almost anyone's blurred face could still be identified.
The researchers said only 10 fully-visible examples of a person's face were needed to identify a blurred image with 91.5 per cent accuracy. With an average of just 1.25 tagged images, the system could still correctly identify an individual 56.8 per cent of the time, which is 73 times higher than chance would allow.
It fared worse in "across event" measurements - those in which the lighting, clothing and context had changed - but still performed at 17 times the rate of chance, identifying people in 31.9 per cent of obscured images.
The researchers also tested different obfuscation methods to see if there was variance in the identification performance. By attempting to recognise non-tagged Gaussian blurs, white boxes and black box images they found that the best method for staying anonymous is to post all your pictures to Facebook with a black box over your face and shoulders. The next safest would be blocking it out with a white box, then a Gaussian blur.
This article was originally published by WIRED UK
