Plaintext attack on joint transform correlation encryption system by convolutional neural network.

Area of Science:

• Optics and Photonics
• Computer Science
• Cryptography

Background:

• Joint transform correlation (JTC) is a popular image encryption technique due to its real-valued ciphertext and relaxed alignment requirements.
• Traditional JTC decryption often relies on phase recovery algorithms, which can be complex and sensitive to errors.

Purpose of the Study:

• To investigate the vulnerability of JTC image encryption systems to deep learning-based attacks.
• To develop a novel method for direct ciphertext-to-plaintext conversion using neural networks.

Main Methods:

• A convolutional neural network (CNN) was trained on a large dataset of JTC ciphertexts and their corresponding plaintexts.
• The CNN architecture incorporated sigmoid activation functions and dropout layers for enhanced speed and accuracy.
• The trained CNN effectively learned to simulate the encryption key.

Main Results:

• The trained CNN model demonstrated the ability to directly convert JTC ciphertexts into their original plaintexts.
• The neural network-based approach proved more efficient than traditional phase recovery methods.
• The trained equivalent key exhibited notable robustness.

Conclusions:

• JTC image encryption systems are susceptible to sophisticated deep learning attacks.
• CNNs offer a powerful alternative for breaking JTC encryption, providing direct decryption capabilities.
• The proposed neural network method is feasible and effective, as confirmed by computer simulations.