05 Feb 2025
Combination of holograms and AI enhances security for currencies, healthcare, communications.
Researchers at Greece's University of Crete and the Foundation for Research and Technology Hellas (FORTH) have developed an optical encryption approach said to achieve an exceptional level of security.As described in Optica, the system uses trained neural networks to retrieve elaborately scrambled information originating from a hologram, decoding the intricate spatial information in the scrambled images.
"From rapidly evolving digital currencies to governance, healthcare, communications and social networks, the demand for robust protection systems to combat digital fraud continues to grow," said project leader Stelios Tzortzakis.
"Our new system achieves an exceptional level of encryption by utilizing a neural network to generate the decryption key, which can only be created by the owner of the encryption system."
Optical encryption involves securing data at the optical transport level of a network, which avoids slowing the overall system with additional hardware at the non-optical levels instead. This approach can make it easier to implement authentication mechanisms at both ends of the transfer to ensure data are genuine.
The project studied whether optical information, such as the shape of a target, could be encoded in holograms of those shapes and then transferred via ultrashort laser filaments propagating in a highly nonlinear and turbulent media. This leaves the initial data completely scrambled and not retrievable by any experimental or physical modeling system, according to the researchers.
Data scrambled by passage through ethanol liquid
The experimental set-up involved a femtosecond laser passing through a prepared hologram and then proceeding into a cuvette filled with liquid ethanol. Laser filamentation and thermal-induced turbulance in the liquid caused the optical data to emerge as a highly scrambled and randomized image, recorded on a CCD sensor.
"The challenge was figuring out how to decrypt the information," said Tzortzakis. “We came up with the idea of training neural networks to recognize the incredibly fine details of the scrambled light patterns. By creating billions of complex connections, or synapses, within the neural networks, we were able to reconstruct the original light beam shapes.
In trials, the system was applied to the encryption and decryption of thousands of handwritten digits and reference shapes. After optimizing the experimental procedure and training the neural network, the encoded images were accurately retrieved 90 to 95 percent of the time, with further improvements possible through more extensive training of the neural network.
The team is now working on ways to employ a cheaper and less bulky laser system, as a necessary step towards commercializing the approach for a range of potential industrial encryption applications.
"Our study provides a strong foundation for many applications, especially cryptography and secure wireless optical communication, paving the way for next-generation telecommunication technologies," said Tzortzakis.
© 2025 SPIE Europe |
|