Deep-learning-based cross-talk free and high-security compressive encryption with spatially incoherent illumination

Area of Science:

• Optics and Photonics
• Information Security
• Artificial Intelligence

Background:

• Incoherent optical cryptosystems offer advantages in noise immunity and alignment tolerance.
• Compressive encryption is crucial for efficient internet data exchange.
• Existing multi-image encryption methods often suffer from cross-talk noise and linearity issues.

Purpose of the Study:

• To propose a novel optical compressive encryption approach using spatially incoherent illumination.
• To leverage deep learning (DL) and space multiplexing for enhanced security and efficiency.
• To address limitations of current encryption schemes, including cross-talk noise and vulnerability to attacks.

Main Methods:

• Utilizing a scattering-imaging-based encryption (SIBE) scheme for transforming plaintexts into scattering images.
• Employing random sampling and space multiplexing to integrate multiple encrypted images into a single ciphertext.
• Applying deep learning (DL) to solve the ill-posed problem of recovering scattering images from sampled data during decryption.

Main Results:

• The proposed method effectively encrypts and decrypts multiple images without cross-talk noise.
• Deep learning successfully resolves the ill-posed decryption problem, demonstrating high recovery accuracy.
• The system shows robustness against ciphertext-only attacks due to the elimination of linearity.

Conclusions:

• The novel incoherent optical compressive encryption approach is effective and feasible.
• The integration of DL significantly enhances the security and performance of optical encryption.
• This method provides a promising solution for secure and efficient data exchange via the internet.