Research On Medical Image Digital Watermarking Algorithm Based On Deep Learning

With the development of medical image sharing and remote diagnosis technology,more and more medical images need to be transmitted on the network.In the process of network transmission,medical images face security risks such as illegal copying,theft,and leakage.Digital watermark can solve such security risk problems.Therefore,the medical image watermark technology used for copyright protection has become a research hotspot.Aiming at the problem that the traditional medical image watermarking method is not universal,this paper introduces deep learning into the research of digital watermark technology,and adaptively designs a suitable watermark algorithm according to the characteristics of medical images.The main research contents are summarized as follows:A robust reversible medical image watermarking algorithm based on Deep Residual Network(Res Net)is proposed.Firstly,the algorithm uses the depth residual model to extract the depth feature information of the medical image,adaptively determine the optimal embedding intensity,and balance the invisibility and robustness of the watermark;In addition,the genetic algorithm and fuzzy C-means clustering algorithm is used to divide watermark region dynamically,According to the clustering results,the watermark information is extracted,This method can overcome the influence of signal attack on watermark image and enhance the robustness of reversible watermark algorithm..The experimental results show that the PSNR value of the watermarked medical image can reach more than 36 d B,which has good image quality;The algorithm can completely restore the carrier medical image to meet the reversibility requirement;After the watermarked image is subjected to different types of attacks,the algorithm can still extract watermark information accurately,showing strong robustness.Then,a robust watermark algorithm based on the combination of DCGAN and U-Net is proposed.Firstly,the algorithm uses the advantages of U-Net with reconstructing high-precision medical images and DCGAN with againsting each other to combine U-Net and DCGAN to reconstruct watermarked Diffusion Weighted Imaging(DWI);Secondly,different types of noise attacks are introduced when training the network,so that the network model learns the robust features of the image against noise attacks;Finally,the watermark information is extracted through the encoder part of U-Net.The experimental results show that the PSNR value of the watermarked DWI image can reach more than 39 d B,which meets the high quality requirements of medical images;The watermark is robust to different types of attacks;And the algorithm changes the DWI image dispersion characteristics parameters more small,which does not affect the clinical diagnosis.