Generative Adversarial Network and Chaotic Map-Based Multi-Layer Medical Image Encryption

Area of Science:

• Computer Science
• Information Security
• Medical Imaging

Background:

• Secure management of cryptographic keys is crucial for medical image communication systems.
• Existing methods often face challenges with key storage and transmission efficiency.
• The pre-shared protected-generator model offers a potential solution for overhead reduction.

Purpose of the Study:

• To propose a multi-layer image encryption algorithm for enhanced security and efficiency in medical image communication.
• To reduce per-image key storage and transmission overhead.
• To leverage a pre-shared protected-generator model for key management.

Main Methods:

• Integration of a Generative Adversarial Network (GAN) for key image generation.
• Utilization of a Piecewise Linear Chaotic Map (PWLCM) for chaotic parameters.
• Incorporation of DNA complement operations and bit-level zigzag permutation for diffusion and confusion.
• Generation of key images from image-specific 100-dimensional noise vectors.

Main Results:

• High entropy values (exceeding 7.996 bits) indicate strong randomness.
• Near-zero adjacent pixel correlations demonstrate effective diffusion.
• Robustness against significant data loss (75% cropping) and noise (50% salt-and-pepper).
• High sensitivity to changes, evidenced by NPCR (99.60%) and UACI (33.46%) values.

Conclusions:

• The proposed algorithm offers a secure and efficient solution for medical image encryption.
• The novel key generation method significantly reduces key storage and transmission overhead.
• The algorithm demonstrates strong security, robustness, and competitive performance against existing methods.