We live in false times. Of course, reality is as common as ever, but it’s getting harder and harder to find it.
Perhaps nowhere is this phenomenon more disturbing than in the strange world of “deepfakes” and other computer-generated faces of people who don’t really exist (yet they look like reality).
Now scientists have developed a technique that could help us detect if the faces we are looking at are really genuine people, unlike the ghosts evoked by artificial intelligence (AI).
According to a new prepress study led by the first author and computer scientist Hui Guo of New York State University, the secret is in the eyes, specifically the shapes of the student.
(Guo et al., ArXiv, 2021)
At the top: a breakdown of the ocular anatomy that includes the regular shape of the pupil (at the top), along with a comparison of a real face and the pupils (left), with some artificial (right).
Taking advantage of the artificial eyes of fake faces created by a machine learning system called Generative Adversative Network (GAN), the researchers noticed something funny about the students.
Unlike actual students, many of the forgeries were not really round.
“Students have almost circular shapes for healthy adults,” the team explains in the study.
“Compared to the actual faces, we observe that visible artifacts and inconsistencies can be observed in the eye regions of the faces generated by GAN.”
According to the researchers, this strange gift is due to the fact that GAN models have no knowledge of the anatomy of the human eye, especially with regard to the geometric shapes of regular pupils.
To explore the extent of this revealing phenomenon, the researchers developed a detection tool that automatically extracts students ’contours from the eyes of the photos and evaluates them to see if they have elliptical shapes.
In an experiment that used the tool against a database of 2,000 images (1,000 are real faces and 1,000 are fakes), the system worked reliably to distinguish the two groups.
“We have widely found irregular pupil shapes on faces generated by high-quality StyleGAN, which are different from real human pupils,” the researchers explain.
“We propose a new physiology-based method that can use the irregular shapes of the pupil as a signal to detect faces generated by GAN, which is simple but effective.”
According to the team, technology like this could one day help counteract the malicious use of realistic-looking fakes that are used to trick people into social networking platforms, among other places.
At least until someone can teach AI the right way to be a student, of course.
Findings are available on the prepress website arXiv.org.