From NewScientist
Deepfake makers can now evade an unusual detection method
By Jeremy Hsu
Deepfake videos that feature digital manipulations of people’s facial expressions and voices can also depict realistic heartbeats, making them even harder to spot.
“We now know that just because a person in a video has a measurable pulse, it doesn’t mean that we can assume they are real,” says Hany Farid at the University of California, Berkeley, who was not involved in the research.
This development comes as deepfakes that have been digitally altered or generated by artificial intelligence are ensnaring celebrities and ordinary people alike in convincing but false pornography, financial scams and political propaganda. Previously, researchers had experimented with spotting deepfakes by identifying changes in skin colour related to blood flow and heart rate, but this research shows that some deepfake videos can still present a passable pulse...
Then, the researchers inserted digitally altered faces into their genuine videos – a move that should have alerted their deepfake detector. Instead, they found that the detector perceived realistic pulses in both the fakes and the original videos. Then, the researchers inserted digitally altered faces into their genuine videos – a move that should have alerted their deepfake detector. Instead, they found that the detector perceived realistic pulses in both the fakes and the original videos.
“Just because one or a few deepfake generators can reproduce this physiological signal, it doesn’t mean that all deepfake generators can,” says Farid...
Hany Farid is a professor in the Department of Electrical Engineering & Computer Sciences and the School of Information at UC Berkeley.
