Research On Robust Zero Watermarking Algorithm For Medical Images Based On MobileNet

With the continuous development of information technology,more and more medical images are uploaded to the Internet platform and transmitted through the Internet.This way of transmitting information makes it much easier for doctors to diagnose patients,however,medical images are also easy to be obtained by criminals in the transmission process,and the patient privacy information on the images will be leaked.In addition,pixel information of medical images is very important and should not be changed at will,otherwise it is easy to cause misdiagnosis by doctors.So how to protect the patient privacy information on medical images is an urgent problem to be solved.Digital watermarking is a kind of information hiding technology,and zerowatermark technology is a kind of digital watermarking.Zero-watermark technology does not change the pixel information of the image,so this technology can be used to protect patient privacy in medical images.The research content of thesis is the medical image zero-watermark algorithm based on Mobile Net,which can well protect patient privacy information.The research can be categorized into three specific parts:(1)A medical image zero-watermark algorithm based on MobileNetV2 pre-trained network is studied.Firstly,the inherent features of medical images are extracted by Mobile Net V2 pre-trained neural network,and then the features are associated with the watermark with patient privacy information through zero-watermark technology to achieve the effect of watermark embedding.The experimental data show that the algorithm has a good ability to resist geometric attacks.(2)The study focuses on the investigation of a MobileNetV2 transfer learning-based algorithm for medical images zero-watermarking.Firstly,the Mobile Net V2 pre-trained network is trained on the medical data set by means of transfer learning.Then,the inherent features of medical images are extracted by the trained Mobile Net V2 network.Finally,the watermark embedding effect is achieved by combining the zero-watermark technology.Compared with the pre-training model,the algorithm’s anti-geometric attack ability,anti-conventional attack ability and the discrimination degree of different images are improved to some extent.(3)A zero-watermark algorithm for medical images based on residual network and depthwise separable convolution is studied.First,a convolutional neural network is built based on the residual network and depthwise separable convolution,and then the network is trained on the medical data set.Then,the trained network is used to extract the inherent features of medical images.Finally,zero-watermark technology is combined to achieve the effect of watermark embedding.Experimental data show that the algorithm has good performance in robustness and differentiation of different medical images.