Internet users and TV viewers are well acquainted with the method of hiding the face when, for some reason, it does not need to be shown. Usually the image is blurred, or hidden under large pixels. In particular, such a service is offered to its users by YouTube. However, advances in the field of artificial intelligence have made it possible to penetrate into this "through the looking glass" in order to almost exactly restore the blurred image to the original one.
The developers of this software used the machine learning technique of the program for recognizing faces and objects in hidden images. In the process of software development, experts taught the algorithm to recover not only faces, but even blurry text.
Finally, they outsmarted a service called Preserving Photo Sharing, which encrypts portions of an image to make them look blurry.
To achieve these results, the researchers conducted a machine learning course using images, words, faces, and objects. When the program learned to recognize images with an accuracy of 90%, the machine was offered several options for completely new blurry images. Result - the original image was restored by 80-90%.
Among the opponents of the system are those who are in favor of maintaining confidentiality. Among her supporters are representatives of law enforcement agencies, who need to identify fuzzy, blurry images of intruders from surveillance cameras.