Sign In

Communications of the ACM

ACM TechNews

New Deepfake Detection Tool Should Keep World Leaders Safe--for Now

View as: Print Mobile App Share:
Actual photos of former U.S. President Barack Obama (top row), and deepfakes.

Researchers have developed a new digital forensics technique to against artificial intelligence-doctored "deepfake" videos.

Credit: University of California, Berkeley

University of California, Berkeley, and University of Southern California researchers have developed a new digital forensics technique to protect world leaders and celebrities from artificial intelligence-doctored "deepfake" videos.

The technique uses machine learning to analyze an  individual's style of speech and movement, known as a "softbiometric signature."

The researchers used an existing tool to extract the face and head movements of individuals, and created their own deepfakes for Donald Trump, Barack Obama, Bernie Sanders, Elizabeth Warren, and Hillary Clinton, using generative adversarial networks.

The team then used machine learning to distinguish head and face movements that characterize the real person; these subtle signals are not currently modeled by deepfake algorithms.

During testing, the method was at least 92% accurate in identifying several variations of deepfakes, including face swaps, and cases in which an impersonator used a digital puppet.

From Technology Review
View Full Article


Abstracts Copyright © 2019 SmithBucklin, Washington, DC, USA


No entries found