In 2024, it is almost trivial to create realistic AI-generated images of people, which has led to concerns about how these deceptive images could be detected. Researchers at the University of Hull have recently introduced a new method for detecting fake images generated by artificial intelligence by analyzing reflections in human eyes. The technique, presented last week at the Royal Astronomical Society’s National Astronomy Meeting, modifies the instruments 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 supervision of Dr. Kevin Pimbblet, professor of astrophysics.
Their detection technique is based on a simple principle: A pair of eyes that are illuminated by the same set of light sources will typically have a similarly shaped set of light reflections in each eyeball. Many AI-generated images produced to date do not account for eyeball reflections, so the simulated light reflections are often inconsistent between each eye.
In some ways, an astronomical angle isn’t always necessary for this kind of deepfake detection, as a quick look at a pair of eyes in a photo can reveal inconsistencies in reflection, something portrait artists need to keep in mind. But the application of astronomical tools to automatically measure and quantify eye reflections in deepfakes is a new development.
Automatic detection
In a post on the Royal Astronomical Society blog, Pimbblet explained that Owolabi developed a technique to automatically detect eyeball reflections and indexed the morphological features of the reflections to compare the similarity between the left and right eyeball. Their findings revealed that deepfakes often show differences between a pair of eyes.
The team used methods from astronomy to quantify and compare eyeball reflections. They used the Gini coefficient, which is typically used to measure the distribution of light in images of galaxies, to assess 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 indicates concentrated light in a single pixel.
In a paper to the Royal Astronomical Society, Pimbblet drew a comparison 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 centrally compact, whether they are symmetrical. and how smooth they are, we analyze the distribution of light.”
The researchers also investigated the use of CAS (concentration, asymmetry, smoothness) parameters, another tool from astronomy to measure the galactic distribution of light. However, this method has proven to be less effective in identifying fake eyes.
Detection arms race
While the eye reflection technique offers a potential avenue for detecting AI-generated images, this method may not work if AI models evolve to include physically accurate eye reflections, perhaps applied as a post-image creation step. This technique also requires a clear, close view of the eyeballs to work.
This approach also risks creating false positives, as even authentic photos can sometimes show inconsistent eye reflections due to different lighting conditions or post-processing techniques. However, eye reflection analysis can still be a useful tool in a larger deepfake detection toolkit that also takes into account other factors such as hair texture, anatomy, skin detail and background consistency.
While the technique is promising in the short term, Dr. Pimbblet warned that she was not perfect. “There are false positives and false negatives; you won’t get everything,” he told the Royal Astronomical Society. “But this method gives us a foundation, a plan of attack, in the arms race to detect deep fakes.”