The development of generative artificial intelligence has enabled fake photos to increase to incredible levels. Now deepfake Even the most careful person may have difficulty understanding whether an image is real or not.
Researchers are also working to solve this situation and detect fake photos. A new study is one of them. According to experts, fake images, With a tactic used by astronomers understandable.
Inconsistencies in the eyes reveal that the image is a deepfake
*The image on the left is real, the one on the right is a deepfake.
In the research presented to the Royal Astronomical Society earlier this week, scientists stated that they could integrate a method by examining photographs of galaxies into deepfake images. They will do this by analyzing the eyes.
*These eyes are all fake.
Actually the method is quite simple. Experts say the people in the photos It can be determined whether it is fake or not by the reflection in the pupils of the eyes. says. It is understood that if the reflections in the two eyes match, the photo is likely to be real, and if they do not match, it is probably a deepfake.
It is possible to understand this if we look carefully at the sample images shared and the fake images we encounter in our daily lives. We can see that there is no realistic reflection in the eyes in the deepfake images. Professor Kevin Pimbblet from the University of Hull in the United Kingdom also says the following about the method: “The reflections in the pupils are consistent in the real person, but appear inaccurate in the fake person.”
In the astronomy method that inspired this, researchers take similar approaches to measure the shapes of galaxies and understand them better. They do this by analyzing whether they are symmetrical or not and their light distribution.
It is possible to say that the method of recognizing pupils can be useful in many current deepfake photos. However, considering how fast artificial intelligence is developing, it would not be wrong to predict that these inconsistencies in the eyes will decrease in a few years. Then it is a matter of curiosity whether this method will work or not.