>
Tech giant Intel thinks it has a solution for the growing deepfake problem.
Earlier this week, the company unveiled FakeCatcher, a brand new software solution that takes a new approach to deepfake video analytics. Reportedly, it can recognize deepfake videos with 96% accuracy.
Like previous deepfake analytics solutions, this one leverages the power of machine learning (opens in new tab). Instead of looking for inconsistencies in the video itself, FakeCatcher analyzes the content to determine whether the person in the video is a real human who was ever recorded or a synthetic product.
(In)visible changes in the face
How does it achieve that? According to Ilke Demir, senior staff researcher at Intel Labs, it can tell whether the person in the video has a beating heart or not.
“When our heart pumps blood, our veins change color,” the Intel report says. “These blood flow signals are collected all over the face and algorithms translate these signals into spatio-temporal maps. Then, with the help of deep learning, we can immediately detect whether a video is real or fake.”
The method is also known as photoplethysmography (PPG), a proven way to measure the amount of light that blood vessels in living tissue absorb or reflect.
Speak against Venture Beat (opens in new tab), Demir said the color changes are invisible to the human eye, but not to a computer. “PPG signals are known, but they have not been applied to the deepfake problem before.”
She also explained that FakeCatcher collects PPG signals from 32 different places on the face.
“We take those maps and train a convolutional neural network on top of the PPG maps to classify them as fake and real,” Demir said. “Then, thanks to Intel technologies like [the] Deep Learning Boost framework for inference and Advanced Vector Extensions 512, we can run it in real time and up to 72 concurrent detection streams.”
Demir built FakeCatcher together with Umur Ciftci of the State University of New York in Binghamton. Apparently, deepfakes are a growing concern as the barrier to entry gets lower and it becomes even easier to create highly convincing videos.