A Trusted Medical Image Zero-Watermarking Scheme Based on DCNN and Hyperchaotic System

Area of Science:

• Computer Science
• Medical Imaging
• Cryptography

Background:

• Medical image copyright protection requires lossless methods with high integrity.
• Existing zero-watermarking techniques often prioritize robustness over discriminability analysis.
• Lack of discriminability hinders effective copyright enforcement for medical images.

Purpose of the Study:

• To propose a trusted and robust zero-watermarking scheme for medical images.
• To enhance the discriminability and security of zero-watermarking systems.
• To address limitations in existing methods concerning copyright protection analysis.

Main Methods:

• Utilizing Deep Convolutional Neural Networks (DCNN) to extract feature map matrices from medical images.
• Generating a stable Gram matrix based on inter-channel correlations within feature maps.
• Fusing image features with a trained DCNN and encrypting the zero-watermark using a hyperchaotic system.

Main Results:

• The proposed scheme demonstrates robustness against common and geometric image attacks.
• Experimental validation confirms the scheme's distinguishability, crucial for copyright verification.
• The integration of DCNN and hyperchaotic encryption enhances both security and discriminability.

Conclusions:

• The developed zero-watermarking scheme is suitable for protecting medical image copyrights.
• The method effectively balances robustness and discriminability, outperforming existing approaches.
• The proposed feature evaluation criteria and encryption strategy offer a secure and reliable solution.