Diffractive deep neural network-based chaotic encryption system

Area of Science:

• Optics and Photonics
• Information Security
• Artificial Intelligence

Background:

• Conventional optical encryption methods often face challenges with static phase design and discrete modules.
• There is a need for advanced optical encryption techniques offering robust security and efficient implementation.

Purpose of the Study:

• To propose and validate a novel optical image encryption system.
• To integrate Diffractive Deep Neural Networks (DDNN) with Logistic Chaotic Map (LCM) and 2D Linear Canonical Transform (2D-LCT).
• To overcome limitations of traditional optical encryption systems.

Main Methods:

• Utilizing a Diffractive Deep Neural Network (DDNN) for integrated optical encryption.
• Employing the Logistic Chaotic Map (LCM) for pixel permutation.
• Implementing the two-dimensional Linear Canonical Transform (2D-LCT) for transform-domain diffusion.
• Optimizing diffractive layer phases to embed hybrid encryption into optical processes.

Main Results:

• The proposed system demonstrates high sensitivity to key parameters, enhancing security.
• Simulations confirmed favorable image reconstruction and generalization capabilities.
• The hybrid encryption approach is effectively integrated into optical diffraction and modulation.

Conclusions:

• The developed DDNN-based optical encryption system offers a promising solution for secure image transmission.
• The integration of LCM and 2D-LCT within a DDNN framework proves effective and feasible.
• This approach represents a significant advancement in the field of optical information security.