In the United States, a new tool was introduced that can detect deepfakes by reflections in the eyes with an accuracy of 94%. But it doesn’t work in all cases.
The new AI-based tool offers an easy way to detect deepfakes: the tool processes the light reflected in the eyes and determines whether it is natural or not. The system was created by scientists from the University of Buffalo in the United States. In tests on portrait photographs, the tool was 94% effective.
The system detects fakes by analyzing the cornea, the surface of which reflects the surface in front of the hero of the video. In a photo of a face taken by the camera, the reflection in both eyes will be similar because they see the same thing. But reflections in deepfake videos or images are often different.
The system generates a score that serves as a similarity metric. The lower the score, the more likely the person is a deepfake. The system has shown high efficiency in detecting counterfeits taken from stocks created in the StyleGAN2 architecture. However, the study authors acknowledge that it has several limitations.
The tool’s most obvious flaw is that it relies on a bounced light source in both eyes. Inconsistencies in these patterns can be corrected with manual post-processing, and if one eye is not visible in the image, the method will not work.
It has also only proven effective in portraits. If the face in the image is not looking at the camera, the system is likely to generate false positives.