Research On Medical Image Zero-Watermarking Algorithm Based On Hadamard Transformation And ResNet50

With the continuous updating of modern science and technology,more and more objects tend to be digitized.The traditional medical method in the past has also evolved into today’s digital medical treatment.For digital medicine,it will involve the problem of transmitting many medical images through the network.Due to the inherent security risks of the Internet,medical images are extremely vulnerable to security problems such as information disclosure,illegal downloading,and tampering.In the field of medicine,the protection of patients’ private information is extremely important.Once the patient’s private information is leaked,it is bound to have a negative impact on the patient’s normal life.If the patient information is clearly marked on the medical image,there will be a risk of information disclosure when it is transmitted on the Internet.Therefore,some researchers propose to use digital zero watermarking technology to solve this problem.The invisibility and robustness of digital zero watermark are used to hide the personal information of patients in medical images to ensure their safe transmission on the Internet.Based on the above,thesis presents a research strategy for medical image zerowatermarking algorithms based on Hadamard transform and ResNet50.The programme is as follows:(1)The first research content is to study the zero-watermark algorithm of medical image based on Hadamard transformation.In the first step,the Hadamard-DCT combination transform is used to extract the image features,and then the 64-bit lowfrequency coefficients in the feature region are selected and combined with the perceptual hash algorithm to generate 64 feature vectors.The original watermarking information is scrambled and encrypted by Logistic map chaotic technology to generate the encrypted binary sequence,then,by adopting the zero-watermarking technology,the watermark embedding and extraction process can be effectively realized.The results confirm that the algorithm is robust and can resist several common types of conventional and geometric attacks.(2)The second research content is to study the zero-watermark algorithm of medical image based on ResNet50.Firstly,by using the pre-trained ResNet50 network,the depth features of the image can be extracted and then processed by DCT transformation to obtain the feature matrix of the image.The 64-bit coefficient of the low-frequency part of the matrix is intercepted and the 64-bit feature vectors are generated with perceptual hash algorithm.The Logistic chaotic technique is used to encrypt the original watermarking information to obtain the encrypted binary sequence,and by adopting the zerowatermarking technology,the watermark embedding and extraction process can be effectively realized.The results confirmed that project is invisible and robust,and performance of the algorithm against attacks is better than that of the algorithm in the previous chapter.(3)The second research content is to study the zero-watermark algorithm of medical image based on ResNet50 and transfer learning.Firstly,Fine-tuning the residual network model ResNet50.The medical image dataset is used to train the fine-tuned ResNet50,and then the trained network is used to extract image features,and the features are transformed by DCT to obtain the image feature matrix.In this matrix,64-bit coefficient of the lowfrequency part of the matrix is intercepted as the feature coefficients,and the 64-bit feature vectors are generated with the perceptual hash algorithm.The Logistic chaos technology is used to encrypt the watermark,and then it is combined with zero watermarking technology to realize the whole implementation process of watermarking algorithm.The results confirm that the proposed algorithm has a good robustness.