Researchers might have found a way to detect deepfakes

Deepfakes are a going concern at the moment and the technology used to create fake images and video is only getting better.

It’s comforting to know then that researchers have developed an algorithm which is capable of detecting deepfake images with 71 – 95 percent accuracy.

The algorithm was developed in part by professor of electrical and computer engineering at the University of California, Riverside, Amit Roy-Chowdhury.

The prof worked with four others to create the algorithm.

The algorithm works by analysing an image pixel-by-pixel. Having been fed a wide array of deepfakes already, it has already reportedly learned what qualities a deepfake has.

While the algorithm analyses the image it also passes it through a set of encoding filters which allow it to analyse the image on a more holistic level.

Speaking with IEEE Spectrum, Roy-Chowdhury says the algorithm could be a powerful tool in detecting deepfakes but warns that the same tool could be used by ne’er-do-wells.

“I think we have to be careful in anything that has to do with AI and machine learning today,” the prof said.

“We need to understand that the results these systems give are probabilistic. And very often the probabilities are not in the range of 0.98 or 0.99. They’re much lower than that. We should not accept them on blind faith. These are hard problems.”

Despite this, Roy-Chowdhury and the team are working hard to add the ability to detect deepfake video.

While the algorithm falling into the wrong hands is a concern, we can’t simply do nothing about the existence of deepfakes and allow misinformation to spread.

Oh, and before we forget, the image above was grabbed from a video by Corridor Digital where the team created a fake version of Tom Cruise. You can see how the team was able to make this happen in the video below.



About Author


Related News

Subscribe to
our newsletters

Select the newsletter you would like to receive: