Deepfake is an impressive technology, although often used for nefarious purposes. Companies have been working on ways to distinguish real video from altered video for years, and Intel’s new solution seems to be one of the most effective and innovative.
Deepfakes, which typically involve superimposing someone’s face and voice over another person, started gaining attention a few years ago when adult websites began blocking videos that used the technique of adding the faces of famous actresses to pornstars’ bodies.
Since then, DF films have become more and more advanced. There are many apps that allow users to insert friends’ faces into videos, and we’ve also seen an AI-powered process that brings old photos back to life and puts young versions of actors back on screen. Or vice versa – it visually ages an actor, such as Chris Hemsworth in the new series Limitless, produced by Darren Aronofsky.
In addition to being used to create fake revenge porn, it is also used by scammers applying for remote jobs. There is also an application dedicated to the digital removal of women’s clothes. The biggest problem, however, is how the deepfake led to the spread of misinformation – earlier this year, a fake video of the surrender of Ukrainian President Volodymyr Zelensky was circulated on social media.
Organizations including Facebook, the US Department of Defense, Adobe and Google have created tools designed to identify deepfakes. Intel and Intel Labs version, aptly named fakecatcheruses a unique approach: it analyzes blood flow and eye movement. Kind of like an early version of the Voight-Kampff test from Blade Runner.
Instead of using a method that screens video for warning signs, Intel’s platform uses deep learning to analyze subtle changes in facial color caused by blood flowing through the veins, a process called photoplethysmography, or PPG. Pulse oximeters that measure blood oxygen saturation work on a similar principle.
FakeCatcher analyzes the blood flow in pixels of the image, which is something that deepfake technology has not yet mastered, and analyzes signals from multiple frames. It then passes the signatures through the classifier. The classifier determines whether a given video is real or fake.
Intel claims that in this way, within milliseconds and with 96% accuracy, you can determine whether a video is real. The company added that the platform uses third-generation Xeon scalable processors with the ability to handle up to 72 simultaneous detection streams and runs through a web interface.
A real-time solution with such a high accuracy rate can make a huge difference in the online war against disinformation. On the other hand, it can also make deepfakes even more realistic, as the creators will try to cheat the system.