Cryptanalysis of equal-modulus-decomposition-based optical asymmetric cryptosystems using physics-driven deep learning

doi:10.1016/j.optcom.2025.132440

Cryptanalysis of equal-modulus-decomposition-based optical asymmetric cryptosystems using physics-driven deep learning

2025-09-01

0

PRE

AI

DOI：10.1016/j.optcom.2025.132440

Original

Original request for help

Save

Abstract

En 中文

Optical encryption based on equal modulus decomposition (EMD) is one of the most widely studied optical asymmetric cryptosystems, which can effectively distinguish between the encryption and decryption keys through vector decomposition. However, despite the development of security-enhanced schemes, security risks persist. In this paper, we propose a novel cryptanalysis method for EMD-based cryptosystems using physicsdriven deep learning. Our method trains an attack neural network using the EMD encryption process itself, eliminating the need for large amounts of plaintext-ciphertext pairs. By inputting the ciphertext into the neural network and continuously optimizing the parameters based on the error between the re-encrypted and original ciphertexts, we show that the plaintext can be accurately recovered from the ciphertext alone, without the private key. The proposed physics-driven learning-based attack method is applicable not only to the basic EMDbased cryptosystem but also to security-enhanced schemes, providing a flexible approach for evaluating the security of EMD-based optical cryptosystems.

Keywords:

Optical encryption

Equal modulus decomposition

Optical cryptanalysis

Physics-driven deep learning

Cryptanalysis of equal-modulus-decomposition-based optical asymmetric cryptosystems using physics-driven deep learning

Abstract

Journal

Organization