New Key to Detecting Deepfake Faces: Eyes

A key new method will now allow us to know when a face is or isn’t safe deepfakethat is, to manipulate some type Generative AI.

And the most amazing thing is that this is done using the same methods that astronomers use to analyze observations of galaxies.

New research presented at the Royal Astronomical Society’s National Astronomy Meeting in Hull suggests that AI-generated deepfakes could be detected by analysing human eyes in the same way astronomers study images of galaxies.

A team led by University of Hull master’s student Adejumoke Owolabi found that the light reflections in the eyes of deeply fake people simply did not match.

“The reflections in the eyeballs are consistent with the real person but are incorrect (physically) for the fake person,” Kevin Pimbblett, an astrophysics professor at the University of Hull, said in a statement.

The researchers analyzed light reflections in people’s eyeballs in real and AI-generated images. They then used methods commonly used in astronomy to quantify the reflections and checked for consistency between the left and right eyeball reflections.

In fake images, there is often a lack of uniformity in the reflections between each eye, while in real images, the reflections are usually the same in both eyes.

“To measure the shape of galaxies, we look at whether they are compact at the centre, whether they are symmetrical and how smooth they are. We look at the distribution of light,” said Professor Pimplet.

“We identify reflexes automatically and analyze their morphological characteristics using CAS indices. [concentración, asimetría, suavidad] and Genie to compare the similarity of the left and right eyeballs.

The Gini coefficient measures the distribution of light in any image of a galaxy. Sort the pixels by their brightness and compare the results to a perfectly uniform distribution.

“It’s important to note that this is not a panacea for detecting fake images,” Professor Pimbblett added. “There are false positives and false negatives; you’re not going to get all of that. “But this method gives us a framework, a plan of attack, in the arms race of detecting deepfakes.”