Research On Robust Watermarking Algorithm For Diffusion Weighted Image

With the development of remote medical technology,medical institutions provide remote services through digital medical images.However,unprotected images may result in patient information leakage and copyright disputes.Digital watermark algorithms can embed watermark information to protect image integrity and ownership.However,there are few watermark algorithms for high-dimensional medical images,especially Diffusion-Weighted Imaging(DWI)images.DWI technology has significant value in disease diagnosis and treatment,and developing a digital watermark algorithm for DWI images is crucial for protecting medical image copyrights.Traditional robust lossless watermark algorithms require targeted design,while deep learning-based watermark algorithms do not require manual design but learn features from a large amount of data and adaptively adjust.This makes deep learning-based watermark algorithms more adaptable to different types of medical images,with better universality and flexibility.Therefore,deep learning watermark algorithms can effectively protect medical image copyrights and maintain medical information security and sustainable development of the medical industry.To address these issues,this paper proposes two deep learning-based robust lossless watermark algorithms for DWI images:(1)A robust zero-watermark algorithm for DWI images based on multiscale feature fusion.A Siamese network that fuses multiscale features and attention mechanisms is proposed to adaptively extract robust features of DWI images and incorporate three prior domain knowledge features into the network to enhance watermark distinguishability,while reducing watermark storage capacity through key slice selection.(2)A two-stage robust reversible watermark algorithm for DWI images based on deep learning.The first stage is pre-training of the robust watermark network.The second stage reversibly embeds the difference image between the watermarked image and the carrier image as compensation information into the robust watermark image,enabling blind recovery of the carrier image without attack,while extracting the robust watermark to complete copyright authentication when under attack.