Spotting fake videos with a 96% accuracy rate, Intel’s FakeCatcher deepfake detector has joined the war on fake news.
The rise of convincing deepfake videos, literally putting fake words in people’s mouths, has become the scourge of the internet – with the potential to destroy reputations, shape public opinion and even influence the outcome of elections.
As deepfake videos become easier to create and harder to spot, they have triggered a technological arms race in the fight to detect them and ensure the public is not deceived.
Intel’s new FakeCatcher deepfake detector can identify otherwise convincing fake videos in milliseconds.
Most deep learning-based deepfake detectors look at raw data, in an attempt to spot signs of inauthenticity and identify what is wrong with a video. Intel’s FakeCatcher takes a different approach, instead looking for signs of authenticity by assessing what makes us human.
FakeCatcher picks up on subtle clues such as the blood flow in the face. By studying the slight changes in the colour of the veins, it can instantly detect whether a video is real or fake. The technology also takes advantage of eye gaze-based detection.
There are several potential use cases for FakeCatcher, according to Intel. Social media platforms could leverage the technology to prevent users from uploading harmful deepfake videos. Meanwhile, global news organisations could use the deepfake detector to avoid inadvertently sharing and amplifying manipulated videos. Nonprofit organisations could also employ FakeCatcher to democratise detection of deepfakes for everyone.
FakeCatcher was developed by Intel Labs researcher Ilke Demir and Umur Ciftci from the State University of New York at Binghamton.
“Deepfake videos are everywhere now. You have probably already seen them; videos of celebrities doing or saying things they never actually did,” Demir says.
“FakeCatcher is the first approach that is telling us why we are real, why we are authentic, it’s the watermark of being human.”
More on eSafety at GadgetGuy.
