New Faceless Recognition Tech Doesn’t Care If You Hide Your Mug Or Not
Once it has 10 photos of you, this system is 91.5 percent accurate
The increased use of facial recognition is worrisome enough as is, but what’s even more unsettling is the fact that people are working on a body recognition technology that barely needs to see your face to work.
The Max Planck Institute in Saarbrücken, Germany recently published a paper on a “Faceless Recognition System,” which trains a neural network to analyze images of a person with their face both blurred and visible. Afterwards, the system is able to identify a person using non-facial features, like stance and surroundings.
Motherboard reports that even when there were just 1.25 examples of someone’s visible face in a photoset, the FRS could correctly ID the person 69.6 percent of the time. Of course, the more clear images the FRS had to learn from, the better it could perform. When there were 10 examples of a person’s visible face, the FRS was right 91.5 percent of the time. This means as long as the system has a few pictures of your face, it could probably figure out who you are. That raises privacy issues, of course, and the researchers are well aware.
“From a privacy perspective, the results presented here should raise concern,” the report states. “It is very probable that undisclosed systems similar to the ones described here already operate online. We believe it is the responsibility of the computer vision community to quantify, and disseminate the privacy implications of the images users share online.”
And when you consider that this is a relatively small research institute working on this project, it makes you wonder what bigger companies, like Facebook, are up to. Facebook has previously claimed that its facial recognition tech is accurate 83 percent of the time, which is already pretty damn good. So for those who would prefer to stay away from the watchful eye of the internet, thinks might just get a little bit harder.