It is said that the eyes are the windows to the soul. Now, researchers claim that they can also serve as a reliable means of detecting deepfakes. To achieve this, one must utilize tools originally designed for studying galaxies and apply them to the human iris. The way light reflects in the eyes holds the key, explains Adejumoke Owolabi, a master's student at the University of Hull in the UK.
Owolabi collaborated with Kevin Pimbblet, a professor of astrophysics and director of the Centre for Advanced Studies in Data Science, AI, and Modelling. Together, they compared authentic images of individuals with those altered by AI-generated deepfakes. To analyze the light reflection in the eyes of both groups, the team employed two astronomical tools: the Gini coefficient and the CAS system.
The Gini coefficient measures the concentration of light in galaxy images based on pixels. Depending on whether the Gini value is 0 or 1, researchers can ascertain whether the structure of a galaxy is smooth or clustered, indicating elliptical or spiral galaxy shapes, respectively. Similarly, the CAS system allows astronomers to measure light distribution in galaxies to determine their morphology.
The team applied both tools to compare the left and right eye irises in both real and AI-generated images. Though the CAS system proved to be insufficient in effectively detecting deepfakes, the Gini coefficient revealed a significant difference.
If the light reflections in both eyes match, the photo is likely authentic. If not, the results point towards a deepfake.
"It is important to note that this is not an infallible method for detecting fake images. There are false positives and false negatives. But this method provides us with a foundation, a plan of attack in the arms race to detect deepfakes," said Professor Pimbblet.
Source: thenextweb
Comments (0)
There are no comments for now