How do you spot a deepfake image of a person? The answer might be to look into their eyes.
That is according to astronomers at the University of Hull in the UK who say that AI-generated pictures can be unmasked by analyzing human eyes in the same way that astronomers study images of galaxies.
The team analysed reflections of light on the eyeballs of people in real and AI-generated images.
They then employed methods typically used in astronomy to quantify the morphological features of the reflections in both eyes.
“To measure the shapes of galaxies, we analyse whether they’re centrally compact, whether they’re symmetric, and how smooth they are,” notes Hull astrophysicist Kevin Pimbblet. “We analyse the light distribution.”
They found that fake images often lacked consistency in the reflections between each eye, whereas real images generally show the same reflections in both eyes.
Yet Pimbblet warns that the technique is “not a silver bullet” when it comes to detecting fake images.
“There are false positives and false negatives [so] it’s not going to get everything,” he adds. “But this method provides us with a basis, a plan of attack, in the arms race to detect deepfakes.”
The work was presented this week at the Royal Astronomical Society’s National Astronomy Meeting in Hull.
