News Release

Light just got harder to hack

Peer-Reviewed Publication

Light Publishing Center, Changchun Institute of Optics, Fine Mechanics And Physics, CAS

As a carrier of information, light possesses many unique features that can be processed to secure data. The way that light bends and moves can be built into devices for essential anti-counterfeit measures in documents, currency, and credit cards. Similarly, scattering light through a material can create a unique pattern, like a fingerprint. Such a system would increase security in many fields, in conjunction with the appropriate authentication, encryption, and verification systems in place. Even better is that any system would be nearly impossible to brute-force due to the enormous length of the key.


However, recent studies have shown that the existing systems have security issues. Suppose someone can obtain both the cyphertext and the corresponding plaintext. In that case, it is easy to break the entire security system and gain all the pass-keys. This problem arises due to the linear nature of the encryption modality. With each pass-key generated in a linear fashion, it is easy to calculate other pass-keys once one is known. It is an enormous security problem for any organization.


In a new paper published in eLight, Professor Guohai Situ and Junfeng Hou from the Shanghai Institute of Optics and Fine Mechanics (Chinese Academy of Sciences) worked together to create a spatial and nonlinear encryption method for images. Their paper, entitled "Image encryption using spatial nonlinear optics," used a photorefractive crystal to generate the interference patterns necessary.


As part of their experiments, the researchers converted a plaintext image to a cyphertext image, using a photorefractive crystal to encode the light via wave mixing. They found that while there were distortions in the resultant image due to imperfections in the crystal, the fundamental outcome proved the principle. It showed that their nonlinear Schrödinger transform was able to encrypt images.


The researchers also tested the encryption method for security purposes. Naturally, any encryption method must be tested to see how difficult it is to crack. If it is too easy to break, it is not worth anything. They found that if nothing but a noise-like pattern is recovered, the nonlinear encryption method is immune to the traditional phase-retrieval-based known-plaintext attack.


They also analyzed the possibility that with machine learning, their proposed model could and should be robust to such an attack. They pointed out that a deep neural network needs a lot of labeled data to crack a system. At the same time, it would be infeasible to generate these data because the proposed nonlinear encryption system is image-dependent.

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.