Israeli researchers led by Yuval Elovitsi of Ben-Gurion University taught a neural network to select makeup to trick facial recognition systems.
Researchers have taught the neural network to deceive facial recognition systems that are installed in public places.
The new development identifies those parts of the face that the recognition system most often reads, then it selects a special make-up that will help deceive the system. Then this makeup is applied to a person: as a result, the accuracy of the face recognition system decreases from 47.5% to 1.2%.
The neural network works in several stages and works with two models for face recognition:
- own (it is also called surrogate);
- target (to be fooled).
First, the algorithm reads several photographs of a person and random people of the same sex. The algorithm then creates a heat map that shows the main areas that are distinctive features of a particular person.
Next, a new face with makeup is created based on the heatmap. It is again passed through the recognition system until it ceases to recognize the person. After that, the resulting makeup can be applied.
The authors note that their development not only deceives recognition systems, but is also invisible to people, since the makeup is normal and does not stand out from the crowd.