A watermarking framework for encrypted medical images via HC chaotic system and deep learning

With the deep iteration and innovation of information technology, medical technology is moving towards informatization and intelligence. This has led to a large-scale collection of medical imaging data that carries patient identification information being stored and disseminated over the network. It greatly increases the risk of medical images being leaked, tampered with, and stolen. To address this issue, a zero-watermarking method for encrypted medical images has been proposed based on HC dual chaos and DWT-ResNet-DCT. Firstly, based on the dynamic characteristic coupling of the Henon chaotic map and the Chen chaotic system, an HC dual-chaotic composite system is innovatively designed. And based on the WHT-DCT transform, it proposes a lossless encryption algorithm characterized by initial value sensitivity and a large key space. While ensuring high encryption efficiency, the algorithm achieves "lossless" decryption of medical images. On this basis, this paper proposes a watermarking algorithm based on DWT-ResNet-DCT for encrypted medical images. This algorithm effectively integrates the characteristics of the DWT transform domain and the convolutional neural network ResNet50, enabling accurate extraction of the feature sequence of encrypted medical images. Finally, experiments verify that the algorithm maintains high NC values (greater than 0.8) under traditional attacks, geometric attacks, and combined attacks, demonstrating excellent anti-attack capabilities, especially having good robustness under high-intensity geometric attacks.