Deep neural network for coded mask cryptographical imaging.

Area of Science:

• Optics and Information Security
• Computational Imaging and Machine Learning

Background:

• Protecting personal privacy during image acquisition is crucial.
• Traditional imaging systems are vulnerable to privacy breaches.
• Novel encryption methods are needed for secure imaging.

Purpose of the Study:

• To propose a novel cryptographic imaging scheme combining optical encryption and computational decryption.
• To prevent privacy breaches during the imaging formation process.
• To explore the potential of deep neural networks in secure imaging.

Main Methods:

• Applied a coded mask for optical encryption of the scene.
• Utilized a deep neural network for computational decryption.
• Employed feature visualization to analyze the encryption process.

Main Results:

• Successfully reconstructed the encrypted image with a Mean Squared Error of 0.028.
• Achieved 100% accuracy in facial expression classification using the Japanese Female Facial Expression dataset.
• Demonstrated that the coded mask synthesizes spatial fidelity while preserving recognition features.

Conclusions:

• The proposed cryptographic imaging scheme effectively balances privacy protection and image utility.
• Deep neural networks offer a powerful tool for computational decryption in secure imaging.
• This framework opens new avenues for unconventional imaging systems.