Even when covering your face, this new artificial intelligence can still recognize who you are

Although this is the advancement of artificial intelligence, it makes the privacy of users more fragile than ever.

The concept of growing popularity among law enforcement agencies, advertisers, and even churches, is that face recognition systems, no doubt, are is becoming one of the biggest threats to privacy.

Scientists have unveiled a new identity system, however dimmed it can recognize the original shape.

The ability to identify any person instantly just by seeing their faces, has created a tremendous power imbalance, with serious consequences that may occur for activities. Free speech and political resistance. More worrisome recently, researchers have demonstrated an identity system, so that even if the face is obscured or obscured, trained algorithms can still identify that person. by matching it with previous models, when comparing the environment around their head and body.

In a new document uploaded to ArXiv server, researchers at Max Planck Academy in Saarbrucken, Germany demonstrated a method of identifying individuals even though most of their images were not attached. Good tag is obscured.

The researchers' system, called "Faceless Recognition System" , trains an artificial neural network on a set of photos containing faces and photos. obscured face. They then used this knowledge to predict the identity of obscured faces by looking for similarities in the areas around the person's head and body.

This system is called "Faceless Recognition System".

The accuracy of this system varies depending on how many faces appear in the set of photos. Even if only 1.25 instances of an individual's face appear completely in photos, the system can still identify the obscured face with an accuracy of 69.6%. If there are 10 instances of an individual's face appearing, the accuracy will increase to 91.5%.

In other words, even if you have done everything to obscure your face in most Instagram photos, the system still has the ability to determine if it is you, as long as you have 1 or In some cases, your face appears fully in a photo.

But identifying this identity can become much more difficult when using a set of photos from "a variety of events" , or elements such as light shining on people's faces and clothes. in the image change.

Identifying this identity can become much more difficult when using a set of photos from "a variety of different events".

The researchers found that when identifying faces obscured by black squares on different events, the system's performance decreased significantly from 47.4% to 14.7%. Even so, the researchers note that this figure is still more than three times more accurate than the "naive" method of identifying obscured faces through guessing.

In the past, Facebook has shown the power of its face recognition algorithm, when it can predict users' identities when they obscure their faces with 83% accuracy, using Use signs such as their posture and body shape. But the researchers say their system is the first system to be trained by making full use of the signs around the blurred and blackened faces.

"From the perspective of privacy, this result further raises concerns," the researchers wrote in their document. "It is possible that the system has not been disclosed, similar to the system described in this document, ready to operate online. We believe it is the responsibility of the computer vision community to identify, and disseminate the link between privacy and user-shared images online ".