Indian-Origin Researcher Has Found Solution To Identify Video Manipulated By Deepfakes AI

There was a time when you could safely say that you'd believe something when you saw it. But thanks to AI editing software like Deepfakes, even your eyes often aren't enough to tell real from fake.

Luckily, a Indian-origin researcher from the US might have just the solution.

Deepfakes are hugely problematic. When it first launched, it was being used by people to swap other people's faces onto the bodies of pornstars in adult clips. Then its usage spread, to editing out facial hair, and eventually even potentially creating fake news clips.

Luckily for us, all of these applications of AI to edit video leaves a sort of fingerprint, one that Amit Roy-Chowdhury from the Video Computing Group at the University of California thinks can be used to identify it.

He's developed a neural network that he's trained to identify manipulated images at the pixel level. "We trained the system to distinguish between manipulated and non-manipulated images and now if you give it a new image, it is able to provide a probability that that image is manipulated or not, and to localize the region of the image where the manipulation occurred," said Roy-Chowdhury.

His AI does this by analysing the borders of objects in an image. When something has been digitally added or removed, the boundary will be of a different quality than of other objects in the image. And though that's the sort of thing you could probably smoothen enough with Photoshop to make invisible to the naked eye, you can't hide it from an AI.

Better yet, Roy-Chowdhury believes that the same principle can be applied to video edits. "If you can understand the characteristics in a still image, in a video it's basically just putting still images together one after another," he said. "The more fundamental challenge is probably figuring out whether a frame in a video is manipulated or not."