Adobe's artificial intelligence is learning to undo retouch in photos

Fake videos and photos are increasinglyare used by cybercriminals to deceive people, and this problem is very troubling to Adobe, which is the developer of popular graphic editors. At the moment, she, together with scientists from the University of California at Berkeley, is developing artificial intelligence to recognize fake photos and cancel all changes made to them - ultimately, the developers want to get the original picture. The prototype tool for undoing retouching has already been created, and already shows interesting results.

The tool is based on the work of the neural network,which clearly sees the differences between edited and original photos. To train it, the developers used a database with paired photos, one of which has no corrections, and the other is edited using the Photoshop Liquify tool to deform the faces.

Neural Networking Photoshop Tool

This type of photo change to developers wasespecially interesting, because a couple of mouse clicks can completely change the expression of a person's face. By editing the photos in this way, fraudsters can easily mislead people, and this has been proven during the experiment. Researchers gathered volunteers and asked them to guess whether the picture was in front of them or not. Only 53% of people showed special vigilance, and the rest did not find anything suspicious. Adobe's artificial intelligence, in turn, recognized the fakes with 99% accuracy.

The developers claim that artificial intelligencemay even show how to edit fakes to get them back to their original state. However, at the moment his recommendations give mixed results.

The idea of ​​the magic universal button "cancel"to return the image to its original state is still far from reality. But we live in a world where it is becoming increasingly difficult to trust the digital information we use, and I look forward to exploring this area of ​​research further, ”said Richard Zhang, an Adobe employee.

Stop spreading fake news is extremely difficult.

This is probably the first study of its kindDuring which scientists want to create a tool for identifying fake photos. Unfortunately, even if such a tool is created, it will not be able to eradicate the distribution of false news - even if it is exposed, the fake can still get into social networks and make people believe the wrong information.

Although we are proud of the influence that Photoshopand other Adobe products have been to the world, we also recognize the ethical implications of our technologies. Fake content is a serious and increasingly pressing issue, Adobe shared on its blog.

Cheating on social networks with fake photos

One of the first cases of cheating usersInternet using fake photos was described by the Associated Press. In the social network LinkedIn a profile of a girl named Katie Jones was discovered - few were able to determine that a potential spy was hidden under this account. However, it turned out to be quite easy to recognize a fake photo - it was created using a neural network and had features that you can read about in our material.

What do you think, are you able to determine whether this is a real photograph, or a fake one? You can talk about this topic in the comments, or in our Telegram-chat.