The facial recognition technology, no doubt, is useful and necessary development, but for all its virtues, it can bear the harm, violating a person’s freedom and allowing spying on him. For this purpose, there was created an algorithm that breaks the system of recognition, not allowing her to invade private life.
The new system is built on the basis of deep machine learning, and therefore able to cultivate in the future. The algorithm uses a dynamic world recognition algorithms. To develop a system of responses, the researchers used two AI that fought with each other. One worked to identify individuals, and the other tried to break it. The result of this war was the creation of a special filter (still images only), which replaces the specific pixels on the images so that the human eye cannot see the difference, but the facial recognition system fails. As stated by the author of the work Professor Parham, Aarabi,
“Now the algorithms of neural networks are constantly evolving and one image can learn a lot about the man. And because of this, privacy is a real problem.”
Professor Aarabi tested his invention on photos 600 people of various ethnic and gender identity. In the end it turned out that the system is able to reduce the proportion of detected individuals from 100% to 0.5%. But that’s not all. During the research was discovered quite a nice bonus: the system does not simply prevent the algorithm to recognize faces, but does not read the color of skin, ethnicity, gender, facial expressions and so on, making the picture “just a picture” for the recognition algorithm.
Created an algorithm that prevents facial recognition system
Vladimir Kuznetsov