Intel has recently launched a technology called FakeCatcher that detects deepfake videos in real-time with high accuracy. They built the technology on algorithms and architectures that use artificial intelligence to detect deepfakes in videos or synthetic content. The tech giant has already incorporated the technology into their Xenon Scalable Processors.
With the rising deepfakes concerns and efforts to combat the widespread misinformation, the research work at Intel started a while ago. In 2020, one of Intel’s research scientists, Ilke Demir, and Binghamton University Professor Umur Ciftc developed the FakeCatcher algorithm. The algorithm detects deepfakes in videos by measuring the biological signals of people.
Converting biological signals into maps
FakeCatcher is built on a principal biological phenomenon called photoplethysmography (PPG). The phenomenon explains that there are changes that happen on the person’s skin when the heart pumps out blood. The algorithm uses a person’s face to detect these changes. PPG can be observed when the veins change color while it pumps out blood.
The algorithm takes the PPG signals from many places on a person’s face and converts them into spatiotemporal maps. From there, a deep learning program analyzes the data and assesses the blood flow as pixels in a video. As per the developers of FakeCatcher, fabricated videos cannot preserve biological signals. This means that it’s possible to use this technology to detect the authenticity of a video.
FakeCatcher is able to separate deepfakes from real videos by learning from big chunks of fake and real data. The algorithm can also determine what generative model was used for creating a particular deepfake video. The algorithm has shown 96% accuracy in detecting synthetic videos and 93.39 percent accuracy in detecting the source model.
The FakeCatcher technology also comes equipped with computer vision and can run up to 72 different detection streams at a time on Intel’s third-generation Xenon scalable processors.
YouTube: Certified human: How new Intel tech detects deepfakes in real-time | Intel
Photo credit: The feature image is owned by Intel and has been provided for press usage.
Sources: Cornell University