No, the pilot did not hold his selfie stick out of a flying plane and photographed himself. There's never been a shark swimming on a highway to Houston, Texas. And the blue moon melon is also just an image processing product. But the pictures are so well manipulated that they can hardly be recognized as fake as an average user.
Ironically, the image manipulation experts of Adobe are working on a software that will detect such altered photos using artificial intelligence (AI). At a conference in Los Angeles, developers from the US company showed how the "About Face" project will in future be used for detective work in image montages.
AI researcher Richard Zhang used portraits to show how the software itself can detect minimal cosmetic corrections. Using the slider, he also blended in a kind of thermal image that showed in which areas the photo had been manipulated. After such an analysis, the software is even able to undo the manipulations, ie to restore the original image.
According to Zhang, the AI has been trained to see if pixels have been dropped from the original, meaning that a picture area has been compressed, such as when a face is slimmed down. In addition, the software should recognize, for example, a mouse click-enlarged forehead, since pixels must be artificially added for this process. Both methods can leave traces that are invisible to the naked eye, Zhang said in his lecture. However, the AI trained in the About Face project could identify and identify these changes.
Whether and when the software will be available, however, is unclear. At the request of an Adobe spokesman said on Friday, "Which of these tools ultimately finds its way into Adobe's products, can not be answered seriously at this time." The presented projects are still in the development process. From circles of the enterprise it means however that one assumes with "About Face" to be able to offer the software in the coming years commercially.
Counterfeiting with AI is getting better and better
Adobe already uses artificial intelligence in its software programs. The AI should help users, for example, if they want to crop fine structures on a photo, for example hair, or tag pictures with similar motifs. In the future, the technology should also help to automatically insert missing persons into group photos and to erase "ers" from audio files.
The catch on such automatic programs: The fakes are getting better and the danger of abuse grows. When deep-fake videos of politicians emerge, voices can easily be impersonated and texts from AI software are written almost as well as by a human being, then tools that can expose such counterfeits are becoming more and more important.