Untrained neural network for cryptanalysis of a phase-truncated-Fourier-transform-based optical cryptosystem

Optical cryptosystem based on phase-truncated-Fourier-transforms (PTFT) is one of the most interesting optical cryptographic schemes due to its unique mechanism of encryption/decryption. Several optical cryptanalysis methods using iterative phase/amplitude retrieval algorithm or deep learning (DL) have also been proposed to analyze the security risks of a PTFT-based cryptosystem. In this work, we proposed an innovative way to attack a PTFT-based cryptosystem with an untrained neural network (UNN) model, where the parameters are optimized with the help of the physical encryption model of a PTFT-based cryptosystem. The proposed method avoids relying on thousands of training data (plaintext-ciphertext pairs), which is an essential but inconvenient burden in the existing data-driven DL-based attack methods. Therefore, the plaintext could be retrieved with good quality from only one ciphertext without any training process. This novel UNN-based attack strategy will open up a new avenue for optical cryptanalysis. Numerical simulations demonstrate the feasibility and effectiveness of the proposed method. (C) 2021 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement.

Automated summary

This study proposes a novel method to attack a PTFT-based cryptosystem using an untrained neural network (UNN) model, optimized with the physical encryption model of the system. By avoiding the reliance on large training datasets, the proposed approach can retrieve high-quality plaintext from only one ciphertext. The numerical simulations demonstrate the feasibility and effectiveness of this method, offering a new avenue for optical cryptanalysis.