The neural network has learned to interfere with facial recognition
And the result is almost one hundred percent.
Experts from the University of Toronto decided to deal with virtual systems that recognize faces in photographs (in smartphones or the Internet). They developed a tool called Privacy Filter (“Filter privacy”), which makes it difficult to recognize faces, changing minor details in the photo.
AI researchers developed a privacy filter to fight facial recognition – https://t.co/PKmFR1nSC6 #Tech pic.twitter.com/eJfQ0qArSV
— Techristic (@techristic) June 1, 2018
In this program, scientists have used two neural networks. First trained to learn the face, and the second — practically invisible to change the image to prevent identification.
Known recognition systems are based on the fact that identify key points on the face (e.g. corners of the eyes, lips or nose) and to calculate the distance between these points, thus creating a unique picture that, like a fingerprint, corresponds to a specific person. Privacy Filter also finds the base point and alters the individual pixels in these areas.
Experts managed to achieve impressive results in this work, the Privacy Filter to cope with the task in 95% of cases.