Researchers at the University of Hull in England have discovered a potential way to identify AI-generated images of people by analyzing reflections in the eyes. This technique, inspired by astronomy, involves comparing the consistency of light reflections in the eyeballs of images. In real photos, the reflections match up, showing the same patterns such as windows or ceiling lights. However, in fake images, there are often inconsistencies in the reflections due to incorrect physics.
To conduct these comparisons, the researchers developed a computer program to detect and analyze the reflections’ pixel values, which represent the intensity of light at a given pixel. By calculating the Gini index, originally used to measure wealth inequality in societies, the researchers could determine the distribution of light across the image of the eyes. A higher difference in the Gini indices between the left and right eyeballs indicated that the image was likely a fake, with approximately 70 percent of fake images showing a significant difference compared to real images.
Although this technique is not foolproof and cannot definitively identify all deepfakes, it can serve as a valuable tool in detecting AI-generated content. Factors such as blinking or proximity to light sources can still make real images appear fake, emphasizing the need for human oversight in the detection process. As AI technology continues to evolve, incorporating methods like analyzing eye reflections could be a crucial step in combating the spread of deepfakes.
By leveraging techniques from astronomy, researchers are pushing the boundaries of AI detection and creating innovative solutions to address the challenges posed by synthetic media. As deepfake technology becomes increasingly sophisticated, interdisciplinary approaches that combine astronomy, computer science, and image analysis will play a key role in safeguarding the authenticity of visual content in the digital age.
