Research On Medical Image Robust Zero-Watermarking Algorithm Based On Deep Neural Network

With the rapid development of medical informatization,modern medical equipment produces a large number of different types of digital medical images every day.Safely storing and transmitting medical images on the Internet and successfully protecting patients’ privacy is the research focus in the field of medical information today.The robust zerowatermarking algorithm for medical images realizes zero-embedding without destroying the integrity of medical images and resolves the contradiction between robustness and invisibility.Therefore,the robust zero-watermarking algorithm is very suitable for the security protection of medical images.The features extracted by traditional medical image watermarking algorithms are mostly shallow features.Under geometric attacks,the extracted values are unstable,which leads to weak robustness against geometric attacks and insufficient embedding capacity of medical image watermarking algorithms.Moreover,the features extracted by traditional methods are usually single,and it is difficult to express the characteristics of medical images,resulting in poor uniqueness of the algorithm.Aiming at the above problems,this paper combines deep neural networks with the medical image watermarking algorithm to design an appropriate watermarking algorithm according to the characteristics of medical images.The main research contents of this paper are as follows:(1)To address the problems of poor robustness against geometric attacks and insufficient embedding capacity of existing algorithms,a robust zero-watermarking algorithm for medical images based on a depthwise overparameterized VGG(DO-VGG)network is proposed.The proposed algorithm uses the pretrained DO-VGG model to extract high-dimensional robust features of medical images,which is used to construct zerowatermarking.The proposed algorithm utilizes depthwise overparameterized convolution to accelerate the convergence of the network.At the same time,the convolutional attention mechanism module is used to extract the deep features of the image from the two dimensions of channel and space,and enhance the algorithm’s ability to resist common attacks and geometric attacks.In addition,the proposed algorithm uses the improved Logistic chaotic system to encrypt the watermarking image.The zero-watermarking algorithm is used to realize the embedding and extraction of the watermarking,which improves the security of the algorithm while meeting the special requirements of medical images.Experimental results prove that the proposed algorithm can effectively resist common attacks and geometric attacks such as rotation and translation.Compared with other medical image watermarking algorithms,the proposed algorithm improves the watermarking embedding capacity of medical images and has stronger robustness.(2)Aiming at the problems of poor uniqueness and security of existing algorithms,a robust zero-watermarking algorithm for medical images based on the Swin Transformer network is proposed.The proposed algorithm uses the Swin Transformer model to obtain multi-scale deep features of medical images.These features have good differentiation and are highly invariant to geometric attacks such as cropping and translation.To enhance the security of the watermarking image,the Hermite chaotic neural network is used to encrypt the watermarking image.The zero-watermarking algorithm is applied during the two stages of watermarking embedding and extraction to ensure the good invisibility of medical images.The experimental results demonstrate that the proposed algorithm has high robustness and uniqueness,and the normalized correlation coefficient(NC)values are kept above 0.83,which can effectively resist different types of attacks.(3)Based on the trained deep neural network zero-watermarking model,design and develop an intelligent zero-watermarking system,and build client and server platforms.The functions of zero-watermarking embedding and extraction are realized,and the practicability of the algorithm is verified.