Corneal Reflections Tell the Difference

Article: How to spot deepfakes? Look at light reflection in the eyes
Source: University of Buffalo, via Technology Networks
Published: March 10, 2021

This study lies at the intersection between computer science and optics. The slang term deepfake  usually refers to visual media that has been edited to replace the person in the original photo or video with another person, often a public figure, in a way that makes it look authentic; the "deep" in deepfake likely refers to deep learning algorithms. Computer scientists are looking into how the optics of the cornea can help to differentiate authentic photos from deepfakes. Because the cornea is foremost a transparent refracting surface, one often forgets that it is also a reflective surface, a convex mirror producing upright, virtual, minified images of the world. Because the two eyes are seeing and reflecting the same thing, the two images should be similar, for example, in shape and color. This is distinguished from the reflections seen in deepfake photos and videos, which are often generated by combining many images and thus produce corneal reflections that are not similar. By comparing the corneal reflections between the two eyes, the investigators' algorithm was reported to be 94% effective at telling the difference between authentic and deepfake portrait-like photos. In order for the tool to work, however, it requires that the photos have a reflected light source, that the original image not have been edited, and that there are two eyes to compare with one another. Additionally, the algorithm only compares differences at the level of pixels rather than broader shapes. Nonetheless, it is an interesting computer science application of corneal optics.

My rating of this study: ⭐⭐