Intel has developed a tool, dubbed FakeCatcher, capable of detecting deepfakes using physical features, with a reliability of 96%. Above all, it is the first system of its kind to achieve this in real time.
The creators of deepfakes just have to watch out, because Intel now has a formidable way of spotting their faked images almost infallibly! The tech giant announces in a blog post having developed, as part of its Responsible AI program, a new tool called FakeCatcher in order to detect deepfakes in a fraction of a second. These images – photos or videos – which use artificial intelligence to place a face on another face – and therefore reproduce “fake” people – are a real danger on the Internet. If they can be entertaining – like on the Snapchat app or to do play Indiana Jones by actor Chris Pratt – they can also be used for malicious purposes. This is particularly the case with the revenge porn – placing someone’s face in a pornographic photo or video – and spreading fake news – such as the false capitulation of the Ukrainian president. They are becoming more and more convincing as technology develops, so new tools and techniques for detecting them are always more than welcome. And in this area, Intel is taking a big step forward, by developing a technology capable of detecting true from false in a few milliseconds – almost in real time – and with a reliability rate of 96%.
FakeCatcher: instant results accessible to all
Unlike detection tools that rely on raw data, FakeCatcher analyzes a video in real time based on physical details, “genuine clues” who “make us human“. The tool inspects the veins in the face, which change color as the heart pumps blood to them as it beats. It analyzes color fluctuations through a method called photoplethysmography (PPG), which measures the amount of light absorbed or reflected from blood vessels. It collects PPG data from about 32 areas of the face and translates it into spatio-temporal maps. It also analyzes eye movement and glare. Then it uses deep learning – l deep learning in French – to determine if what the video shows is real or fake. This technology is based on a set of software and hardware layers from Intel. It is based on third-generation Xeon Scalable processors – specially created by Intel – to run up to 72 analyzes simultaneously, and is accessible via a simple web interface.
While there are other tools capable of detecting deepfakes, instant processing is a big first. Until now, you had to load the videos on complex platforms and wait several hours before getting the results. This is why the prospects offered by FakeCatcher are immense. Intel is already planning to rely on social networks, media, governments and NGOs as clients in order to detect large-scale deepfakes, and thus better fight against disinformation. This is all the more necessary as social networks are increasingly used to follow the news, especially by the youngest.