Differentiating real photos from artificial intelligence (AI) generated images is essential in this era where creation of AI pictures only takes a few clicks. Recent findings indicate that deepfakes can be identified with an astonishing level of accuracy by considering the reflections in people’s eyes.
Key points
This study, which was presented at the National Astronomy Meeting of the Royal Astronomical Society held in Hull, identifies a simple but effective way to distinguish between genuine and counterfeit photos. The research was carried out by Adejumoke Owolabi, an MSc student at the University of Hull who borrowed techniques employed in astronomy to examine human eyeball reflections.
How does it work? If an image is authentic, the light should reflect on a person’s eyes evenly, hence matching the reflections in both eyes. However, such balanced distribution seldom appears in deepfakes. This means that if these two reflections do not correspond correctly with each other then it could be indicative that we are dealing with a synthetically created image.
Artificial ones have incorrect reflections while those for real people remain constant all along their life time said Kevin Pimbblet who is an astrophysicist professor at University of Hull College according his statement “The reflections within eye balls are consistent between humans but inconsistent among fakes”
To arrive at this finding, scientists compared eye ball reflection shown by individuals depicted in genuine photographs against those exhibited by persons portrayed through AI-produced pictures. They measured and quantified these mirrored areas using astronomical procedures while ensuring there is uniformity between left and right eyes during checks for consistence.
When examining galaxies, astronomers take into account things like symmetry as well as light distribution among others. Coincidentally similar techniques were applied here whereby researchers used Gini coefficient which normally analyses how lights spread across galaxies.A value close to 0 indicates evenness throughout an entire area occupied by stars whereas if it approaches 1 then most part would be concentrated around one point.
According to the team fake eyes were not detected by means of CAS (concentration, asymmetry, smoothness) parameters which are another tool from astronomy they used. While it may not be foolproof, this method provides an important weapon in our arsenal against deepfakes as Professor Pimbblet points out “It’s important to note that this is not a silver bullet for detecting fake images”
“There are false positives and false negatives; it’s not going to get everything. But this method provides us with a basis, a plan of attack, in the arms race to detect deepfakes.” Professor Pimbblet added.
