In 2024, it is almost trivial to create realistic AI-generated images of people, which has led to fears of how these fraudulent images can be detected. Researchers at the University of Hull recently unveiled a new method for detecting fake AI-generated images by analyzing reflections in human eyes. The technique, presented at the Royal Astronomical Society’s National Astronomy Meeting last week, adapts tools used by astronomers to study galaxies to examine the consistency of light reflections in eyeballs.
Adejumoke Owolabi, an MSc student at the University of Hull, led the research under the direction of Dr. Kevin Pimbblet, professor of astrophysics.
Their detection technique is based on a simple principle: A pair of eyes illuminated by the same set of light sources will typically have a set of similarly shaped light reflections in each eyeball. Many AI-generated images created to date do not take eyeball reflections into account, so the simulated light reflections are often not consistent between each eye.
In a way, the astronomy angle isn’t always necessary for this kind of deep forgery detection, because a quick look at a pair of eyes in a photo can reveal reflection inconsistencies, something artists who paint should be aware of. portraits. But applying astronomy tools to measure and automatically measure eye reflections in deepfakes is a new development.
Automated discovery
In a Royal Astronomical Society blog post, Pimbblet explained that Owolabi developed a technique to automatically detect eyeball reflections and described the morphological features of the reflections through indices to compare the similarity between the left and right heads. Their findings revealed that deep fakes often show differences between pairs of eyes.
The team applied methods from astronomy to estimate and compare eyeball reflections. They used the Gini coefficient, commonly used to measure the distribution of light in galaxy images, to estimate the uniformity of reflections across the eye’s pixels. A Gini value closer to 0 indicates evenly distributed light, while a value closer to 1 suggests light concentrated on a single pixel.
In the Royal Astronomical Society post, Pimbblet drew comparisons between how they measured the shape of the eyeball reflection and how they typically measure the shape of galaxies in telescope images: “To measure the shapes of galaxies, we analyze whether they are compact in the center, if they are symmetrical, and how smooth they are we analyze the distribution of light.”
The researchers also explored the use of CAS (concentration, asymmetry, smoothness) parameters, another tool from astronomy for measuring the distribution of galactic light. However, this method proved less effective in identifying false eyes.
A discovery arms race
While the eye reflection technique offers a possible route to AI-generated image detection, the method may not work if AI models evolve to incorporate accurate physical eye reflections, perhaps applied as a subsequent step after generating the image. The technique also requires a clear, close-up view of the eyeball to work.
The approach also risks producing false positives, as even authentic photos can sometimes show inconsistent eye reflections due to different lighting conditions or post-processing techniques. But analyzing eye reflections can still be a useful tool in a larger set of deep fake detection tools that also take into account other factors such as hair texture, anatomy, skin detail and background consistency .
While the technique shows promise in the short term, Dr. Pimbblet cautioned that it’s not perfect. “There are false positives and false negatives; it won’t pick up everything,” he told the Royal Astronomical Society. “But this method gives us a baseline, a plan of attack, in the arms race to detect deep forgeries.”
