Two teams of researchers are attempting to use heartbeat detection to identify deepfake videos. The first comes from Intel and Binghamton University, where researchers have developed technology that is not only able to distinguish a deepfake from an original video, but also what deepfake software was used to generate that particular fake.
The Intel and Binghamton solution is premised on the fact that a human heartbeat will create slight changes in skin color as blood flows through the body, and that current deepfake programs cannot yet mimic those subtle variations. The detection system uses photoplethysmography (PPG) technology to watch for biometric signals in 32 PPG cells on a subject’s face.
The researchers were able to identify specific software programs because each deepfake model leaves a unique signature in its efforts to simulate a real human. They identified deepfakes with 97.3 percent accuracy, and specific generative models with 93.4 percent accuracy.
The other solution has been dubbed DeepRhythm, and it represents a collaboration between the Alibaba Group, Kyushu University, Nanyang Technological University, and Tianjin University. DeepRhythm is similarly based on the idea that deepfake tech cannot match real human heartbeats, and uses visual PPG tech and a heart rhythm motion amplification module to separate authentic videos from convincing fakes.
Both teams indicated that they would like to integrate PPG tech into video authentication systems to prevent spoofing and other forms of fraud. Multiple organizations have warned about the growing threat of deepfakes in the past few years, while social media platforms have tried to crack down on them ahead of the upcoming US election.
Of course, the researchers are not the only teams trying to thwart deep fake technology. Microsoft is currently marketing a new deepfake detection system, and iProov has opened a new Security Operations Centre to help watch for biometric storm attacks.
September 4, 2020 – by Eric Weiss