They say that the eyes are the mirror of the soul. Now, researchers claim that they can also be a reliable means of detecting deepfakes. To do this, they would also need to take the tools used to study galaxies and look into the eyeballs with them. It’s all about how light is reflected in the eyes, says Adedjumoke Owolabi, a master’s student at the University of Hull in the UK.
Ovolabi worked with Kevin Pimblett, Professor of Astrophysics and Director of the Centre of Excellence for Data Science, Artificial Intelligence and Modeling. Together, they compared real images of people with AI-generated dipshots. To analyze the light reflection in the eyes of both groups, the team used two astronomical tools: the Gini coefficient and the CAS system.
The Gini coefficient measures the concentration of light in a pixel-based image of a galaxy. Depending on whether the Gini value is 0 or 1, researchers can determine whether the structure of the galaxy is smooth or crowded, indicating the shape of elliptical or spiral galaxies, respectively. Similarly, the CAS system allows astronomers to measure the distribution of light in galaxies to determine their morphology.
The team used both tools to compare the left and right eyeballs in real and AI-generated images. Although the CAS system was not effective enough in identifying dipfakes, it turned out that the Gini coefficient showed a noticeable difference.
If the light reflections in both eyes match, it is most likely a real photo. If not, the results indicate a dipfake.
«It is important to note that this is not a perfect tool for detecting fake images. There are false positives and false negatives. But this method gives us a framework, a plan of attack in an arms race to detect deepfakes»,” said Prof. Pimblett.
Source: thenextweb