Every time you post a photo on a social network, facial recognition systems learn a little more about you. These algorithms compile data about who you are, where you are and the people you know. And according to researchers from theUniversity of Toronto, in Canada, these algorithms are constantly improving. To counter them, the latter had the idea of creating two artificial intelligences (AI) with the aim of disrupting these facial recognition systems. This work will be published and presented to theIEEE International Workshop from 2018.
Imperceptible changes in the photo
The first AI works to identify faces, and the second to disrupt the first’s facial recognition task. The two are constantly fighting and learning to outsmart the other. The result of this long battle is a filter, similar to the ones you can find on Instagram or Snapchat, and which can be applied to photos to protect private life.
In the illustration photo, the differences between the first and the second photo are imperceptible. Indeed, the algorithm only modified certain pixels (visible on the 3rd image in black and green). He only tackled what the first AI is looking for, like the corner of the eye for example. So, by adjusting the corner of the eyes, it creates very subtle disturbances, but large enough to trick the system. The team hopes to make this privacy filter publicly available, either through an app or a website.
Read also :
Should we publish the images of our children on social networks?
Google takes on British health data
