Researchers at Germany's Darmstadt University of Applied Sciences trained an algorithm to remove facial tattoos to improve facial recognition systems.
The process involved automatically adding ink to 41 images of untattooed faces, with tattoos covering 5% to 25% of the face in the image.
A generative adversarial network (GAN) was trained using these images and was able to remove the tattoos, though it had issues with those covering the entire face.
The GAN-altered images were tested against a facial recognition system, halving the system's error rate when the tattoos were removed by the GAN.
However, Kay Ritchie at the U.K.'s University of Lincoln said the use of digitally edited images to train the algorithm means it may not be as effective with unaltered images.
From New Scientist
View Full Article - May Require Paid Subscription
Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA
No entries found