AI from Adobe may determine the manipulation of faces on the photos

ИИ от Adobe может определять манипуляции с лицами на фотографиях

Although the use of tools like Photoshop to digital imaging is not something new, but thanks to the progress of technology, the problem is substantially simplified. This is particularly disturbing in connection with the creation of tools with which the AI allows you to create fake pictures and make people on the video realistic to say tampered with it or behave in any other way fabrications.

In our time, the fight against fake news and information is a very important task, especially with the speed of data dissemination. So Adobe has announced that its experts Richard Zhang (Richard Zhang) and Oliver Wang (Oliver Wang) joined forces with the staff of the University of California, Berkeley Shen-Yu Wang (Sheng-Yu Wang), Dr. Andrew Owens (Andrew Owens) and Professor Alexei Efros (Efros Alexei A.) with the aim to create a system based on AI is able to accurately detect the manipulation of the faces in the images that were edited using software like Photoshop.

“When Adobe Photoshop was first released in 1990, it was an important step towards the democratization of creativity and expression. Since then, Photoshop has had a profound impact on creativity and, more broadly, on our visual culture. Although we are proud of the impact that Photoshop and other Adobe tools made the world, we also recognize the ethical implications of our technology. The credibility of what we see, is becoming increasingly important in a world where image editing has become widespread – fake materials are a serious and growing problem. Adobe is looking for the most useful and responsible ways to implement new technologies in life – so we are constantly exploring the applicability of new technologies like artificial intelligence (AI) to improve trust and credibility in digital media”, wrote the company in his blog.

Adobe project sponsored by DARPA under the program MediFor, is still at an early stage. He is learning precision of the neural network (CNN), a form of deep learning – which the study was able to identify the person subjected to changes as a result of manipulation. Researchers have created an extensive training set of images, writing a script Photoshop using filter Face Aware Liquify at thousands of random pictures from the Internet.

Interestingly, created in Adobe method proved to be much more perfect than the human eye. If people on average were determined by manipulation of persons using the popular method of Face Aware Liquify only in 53% of cases (that is, little more than a random selection), the algorithm based on AI 99 %. This is a very good result.

Adobe promises to continue working towards the establishment of methods for determining image manipulation, video, sound and documents. The goal is to not only identify the manipulation, but also indicate their nature and, if possible, to undo changes made to the original.

Explorer Adobe Richard Zhang commented on preliminary results: “the Idea of creating magic universal button that allows you to return the edited image to the original appearance, is still far from reality. But we live in a world where it is increasingly difficult to trust the digital information that we consume, and I look forward to further progress in this field of research.”

ИИ от Adobe может определять манипуляции с лицами на фотографиях

ИИ от Adobe может определять манипуляции с лицами на фотографиях

ИИ от Adobe может определять манипуляции с лицами на фотографиях

Share