As deepfake videos become more and more believable, they are a cause for serious concern over their ability to be leveraged to misrepresent things like news stories or politicians. Researchers have come up with a clever new technique for detecting deepfakes that can help fight this problem.
The technique, developed by researchers from Intel and Binghamton University, makes use of a person’s heartbeat to detect the difference between real and fake videos. Photoplethysmography, which measures volumetric changes in peripheral circulation and is often used to monitor heart rate, is the technique leveraged in the method. As your heart pumps blood through your body, your skin changes color by small amounts. These small changes are undetectable by the naked eye, but most cameras can easily pick them out, which can then be translated to a heart rate measurement.
The researchers applied this technique in a tool called FakeCatcher and detailed the results in their paper, “FakeCatcher: Detection of Synthetic Portrait Videos using Biological Signals,” which showed impressive results of accuracy of over 90% on various datasets. This works because deepfakes are often created through manipulations of a collection of still images, which means the resulting video lacks the proper variations in skin color a true video would have. Instead of a relatively stable and predictable heart rate pattern, the deepfake pulse is unstable and unpredictable. Check out the video above for more.