Deepfakes quickly became a problem on the internet. In just a few short years, these videos have evolved into web comedy, false job interviews and even pornography without the consent of the person depicted. As artificial intelligence (AI), technology advances, deepfakes will become more convincing. Some companies are now developing their own techniques to detect them.
Intel’s FakeCatcher does exactly that, with surprising accuracy. claims(Opens a new tab) that FakeCatcher can distinguish videos of real people and deepfakes in milliseconds with 96 percent accuracy. How does it work? How? By searching for the substance that gives us life: blood.
FakeCatcher looks for minuscule blood flow signs in pixels of a possible deepfake. Our veins change colour as our hearts pump the blood. Photoplethysmography, a technique used to monitor blood flow in a person, is based on this phenomenon. FakeCatcher uses PPG to capture blood flow signals across the face of a video subject. The tool then converts those signals into spatial-temporal maps that deep learning algorithms can compare with human PPG activity in order to determine if the video subject is real or not.
Intel engineers created FakeCatcher using several of their own software products, including OpenVINO for the deep-learning interface and OpenCV to detect in real time. FakeCatcher is powered by 3rd Generation Intel Xeon Scalable Processors. Intel claims that this allows the tool to handle up to 72 different detection streams simultaneously. This type of capability could be used to squash conspiracy theories and scams as well as maliciously made adult content, depending on where and how FakeCatcher was deployed. Intel did not specify where FakeCatcher would be used now that the testing is complete, but it mentioned it could appear on news websites, social media sites, or nonprofit websites.
FakeCatcher is Intel’s contribution to the growing movement of stopping deepfakes at their source. Google banned deepfake-training from its Colab service earlier this year due to an increase in misinformation. Two weeks later, the European Union said it would take action against Meta and Twitter as well as other tech giants who failed to make a clear effort to combat fake news.
Intel engineers created FakeCatcher using several of their own software products, including OpenVINO for the deep-learning interface and OpenCV to detect in real time. FakeCatcher is powered by 3rd Generation Intel Xeon Scalable Processors. Intel claims that this allows the tool to handle up to 72 different detection streams simultaneously. This type of capability could be used to squash conspiracy theories and scams as well as maliciously made adult content, depending on where and how FakeCatcher was deployed. Intel did not specify where FakeCatcher would be used now that the testing is complete, but it mentioned it could appear on news websites, social media sites, or nonprofit websites.
