Deepfake is one of the controversial technologies created on the basis of artificial intelligence to create fake and fake photos and videos, which can be a great threat to societies and also used for criminal and immoral purposes. Now, Intel has unveiled the first real-time deep fake detection system called FakeCatcher.
Intel claims the product has a 96% accuracy rate and works by analyzing the “blood flow” in video pixels to deliver the results users want in milliseconds.
Ilke Demir, a senior research scientist at Intel Labs, designed FakeCatcher in collaboration with Amur Ciftchi of the State University of New York at Binghamton. The product uses Intel hardware and software, runs on a server, and gets its user interface through a web-based platform.
How the Face Catcher deepfake detection system works
Unlike many deep learning-based fake detection systems that examine raw data, FakeCatcher focuses on clues within videos. In fact, this system uses the photoplethysmogram (PPG) method, which can measure the amount of light that is absorbed or reflected by blood vessels in living tissue.
Demir told VentureBeat about the method:
“You can’t see it with your eyes, but it’s computationally visible. “PPG signals are known but not yet used for deepfake.”
she explained that in FakeCatcher, PPG signals are collected from 32 points on the face, and then PPG maps are created from the temporal and spectral components.
Demir further explains:
“After receiving these maps, we train a convolutional neural network on top of the PPG maps to classify them as fake or real. Then with the help of Intel technologies like the Deep Learning Boost framework for inference and Advanced Vector Extensions 512, we can run it in real-time with up to 72 simultaneous detection streams.”
As we said, deepfake carries many risks, and having an immediate detection system can be essential. However, Demir explained that the development of this system is still in its early stages.
