Deep Learning Toolbox Now Apparently Includes Ground-Up Glass

Scientists in China, Hong Kong, and Singapore suggest ground-up pieces of glass could help in securely encrypting facial images via a new optical cryptosystem.

The system transmits facial-image data through the glass, producing speckles with apparently random scattered patches of light and dark recorded by camera as a secret message.

The researchers said the process generates keys 17.2 billion bits long; they trained a deep learning neural network to decrypt messages by feeding it 19,800 facial images before and after being sent through a given set of ground glass.

Preliminary tests showed the network could decrypt the images with over 98% accuracy.

Hong Kong Polytechnic University's Puxiang Lai described the system as "fast, low-cost, and easy to integrate with other systems."

From IEEE Spectrum
View Full Article

Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA