Cryptanalysis of an optical cryptosystem with uncertainty quantification in a probabilistic model.

Area of Science:

• Computer Science
• Cryptography
• Machine Learning

Background:

• Double Random Phase Encryption (DRPE) is a widely used optical encryption technique.
• Attacking DRPE is crucial for understanding and improving its security.
• Deep learning methods have shown potential in cryptanalysis but often lack reliability assessment.

Purpose of the Study:

• To propose a modified probabilistic deep learning method for attacking DRPE.
• To introduce uncertainty quantification to optical cryptanalysis for the first time.
• To assess the reliability of deep learning predictions in cryptanalysis.

Main Methods:

• A modified probabilistic deep learning model was developed to learn the conditional distribution of plaintext.
• The model was trained to predict plaintext and quantify the uncertainty associated with each pixel prediction.
• Simulation experiments were conducted to validate the method's effectiveness.

Main Results:

• The proposed probabilistic model achieved high accuracy in predicting plaintexts, demonstrating successful cryptanalysis.
• Uncertainty quantification provided pixel-level reliability information without requiring ground truth.
• The method effectively identified unreliable predictions, mitigating risks associated with deep learning models.

Conclusions:

• The modified probabilistic deep learning method is effective for attacking DRPE.
• Uncertainty quantification is a valuable addition to optical cryptanalysis, enhancing model reliability.
• This work advances the security analysis of optical encryption techniques.