2024-08-09 06:00:04
As artificial intelligence (AI) image creation becomes commonplace, identifying fake visuals, including deepfakes, becomes crucial. Scientists have achieved promising results using astronomical image analysis techniques.
Credit: Adejumoke Owolabi
Researchers have found that analyzing reflections in human eyes could reveal deepfakes, similar to astronomical methods used to study galaxies. The study, presented at the Royal Astronomical Society’s national meeting, highlights the inconsistency of reflections in eyes in AI-generated images.
The researchers used astronomical techniques to analyze these reflections. The reflections in the eyes of real people are symmetrical, unlike those in the fake images.
![](https://i0.wp.com/static.techno-science.net/illustration/Source/STD/2024/08/04/Deepfake-Eyes.jpg?resize=800%2C281&ssl=1)
A series of deepfake eyes showing inconsistent reflections in each eye.
Credit: Adejumoke Owolabi
Kevin Pimbblet, professor of astrophysics at the University of Hull, explains that the study of eye reflections can be done in an automated way thanks to indices of concentration,asymmetry and softness (CAS) and Gini. The results show that deepfakes exhibit notable differences between the two eyes.
The Gini index, used to measure the distribution of light in a galaxy, is applied here to assess the coherence of reflections in the eyes. A high Gini index indicates a concentrated distribution of light, while a low index shows a uniform distribution.
![](https://i0.wp.com/static.techno-science.net/illustration/Source/STD/2024/08/04/Eyes-from-Real-Images.jpg?resize=800%2C331&ssl=1)
A series of real eyes showing largely consistent reflections in both eyes.
Credit: Adejumoke Owolabi
However, CAS parameters, also used for galaxy morphology, have failed to effectively predict false eyes. This method is not foolproof, Pimbblet warns, because it carries risks of false positives and negatives.
While eye reflection analysis is not a perfect solution for detecting deepfakes, it does provide a solid foundation for combating the phenomenon.
1723304078
#eyes #betray