**Finally… The Naked Eye and Astronomy Techniques Expose AI-generated Fakes!**
Astronomy techniques are now being used to analyze light reflections in the eyes of people in AI-generated images.
There’s a saying that the eye is the mirror of the soul, and a new study conducted by researchers at the University of Hull in the UK has shown that the eye can also reveal AI-generated fake images, opening new avenues in the fight against deepfakes.
The idea emerged when Kevin Pimblett, a Professor of Astrophysics at the University of Hull, was examining images generated by AI tools like Midjourney and Stable Diffusion. He noticed that the light reflections in the eyes of people in these images provided strong evidence of their inauthenticity.
This simple observation led him to an important scientific discovery: the possibility of using techniques borrowed from astronomy to analyze light reflections in the eyes of people in images. These techniques proved effective in distinguishing between real faces and those generated by AI tools.
**How Does It Work?**
Deepfake technology relies on AI algorithms to generate fake but realistic-looking images and videos. This is done by training models on vast amounts of visual data. When creating an image of a human face, AI uses its extensive knowledge to build the face from scratch, pixel by pixel, or generates it based on real human faces. However, since real images contain light reflections, AI also adds them, but it cannot replicate them accurately. Here, subtle differences emerge between fake and real images that can be detected, such as variations in light reflection between the eyes.
To confirm this hypothesis, Pimblett enlisted the help of Adejumoke Owolabi, a master’s student at the university, to develop a program that could quickly scan the eyes of individuals in different images to assess the differences between light reflections in the left and right eyes in both real and fake images.
The researchers used a vast collection of real images, totaling 70,000, from Flickr, while the fake images were generated using the website This Person Does Not Exist, which relies on AI techniques to create realistic images of people who do not actually exist.
By comparing thousands of real images with AI-generated fake images, the researchers were able to identify subtle differences in light reflections between the eyes in the fake images. Even in cases where the images appeared completely realistic, such as a man wearing glasses, the lens reflections raised suspicion.
However, these subtle differences are difficult for the naked eye to detect. Therefore, the researchers turned to astronomy techniques, namely the CAS parameters system and the Gini index, to accurately identify these differences by measuring the symmetry and uniformity of light distribution within the eye.
**How Did Astronomy Techniques Help Detect Fake Images?**
The researchers used two different astronomical measurement tools, the CAS parameters system and the Gini index, to compare the reflected light in the eyeballs and accurately detect deepfakes.
The CAS parameters system is a mathematical tool that astronomers have used for decades to study galaxies and analyze their light distribution. This system is based on determining the intensity of light in each pixel within an image of the galaxy, using three main parameters: concentration, asymmetry, and smoothness.
The Gini index, on the other hand, is a statistical measure developed by the Italian Corrado Gini in 1912 to measure income inequality. Astrophysicists use it to analyze light distribution in galaxies.
The index works by comparing the distribution of light intensity in galaxy images to an ideal, equal distribution. A value close to zero indicates uniform light distribution, while a value close to one indicates a high concentration of light in a single area.
When these two tools are used together, astronomers can distinguish between different types of galaxy structures, ranging from spiral galaxies to elliptical galaxies.
Applying these tools to images of faces revealed that real images exhibit precise symmetry in light reflections on the eyes—a symmetry that is difficult for AI to replicate.
By/radwa sherif ✏️✏️📚