Vol. 54, Issue 3, pp. 395-407 (2024)

Abstract

Noise attack is a potential threat to optical cryptosystems because the contaminated ciphertext always yields degraded decrypted result. What is more, such contamination can hardly be eliminated by traditional methods, as the ciphertext itself is also a noise-like image. In this paper, we propose a deep-learning-based approach to deal with this problem. The contaminated ciphertexts, which produce unrecognized decrypted images, can yield high quality ones after being repaired by a deep neural network. We take the diffractive-imaging-based encryption (DIBE) scheme as an example to illustrate our method. Numerical results are presented to show the feasibility and validity of the proposal.