Cryptanalysis of random-phase-encoding-based optical cryptosystem via deep learning.

Area of Science:

• Optics
• Cryptography
• Machine Learning

Background:

• Random Phase Encoding (RPE) is increasingly used for optical image encryption.
• Optical cryptosystems based on RPE have gained attention for their security applications.
• However, the security of these RPE-based systems requires thorough investigation.

Purpose of the Study:

• To investigate the vulnerability of RPE-based optical cryptosystems to chosen-plaintext attacks (CPA).
• To develop and employ a deep neural network (DNN) strategy for decrypting images encrypted with RPE techniques.
• To assess the effectiveness of the DNN approach against both Double RPE (DRPE) and Tripe RPE (TRPE) schemes.

Main Methods:

• A deep neural network (DNN) model was designed and trained to learn the encryption-decryption mechanism.
• The trained DNN was utilized as a decryption system to recover original images.
• Numerical simulations were performed to validate the feasibility and reliability of the proposed attack strategy.

Main Results:

• The study successfully demonstrated that RPE-based optical cryptosystems are susceptible to CPA using a DNN.
• The developed DNN model effectively decrypted images encrypted using both DRPE and TRPE schemes.
• The results showed the capability to reconstruct images not included in the original training dataset.

Conclusions:

• Deep learning strategies pose a significant threat to the security of RPE-based optical cryptosystems.
• The DNN-based decryption method is feasible and reliable for breaking DRPE and TRPE encryption.
• Further research is needed to develop more robust encryption techniques against advanced cryptanalytic methods.