Robust Watermarking Algorithm For Medical Images Based On ShuffleNet Convolutional Neural Network

Digital watermark is an effective means of copyright protection of digital media.It plays an important role in the protection of intellectual property rights.Using the invisible and robustness of digital watermarks,we can use digital watermark technology for privacy protection,such as hiding the patient’s personal information in medical images.In the field of medicine,digital medical images have become the main basis for doctors diagnosed,and at the same time play an important role in clinical teaching and scientific research.Compared with natural images,medical images have both the height similarity of the overall structure and the diversity of details.The processing methods for medical images need to be cautious and reliable.Based on this,we hope that after adding watermark information to medical images,we can return the original carrier medical image without damage.Therefore,medical images have high requirements for the robustness and irresponsibleness of the watermark algorithm,and have important research significance for digital watermark technology for medical images.This article mainly studies the medical image digital water marks algorithm based on the LPT-DCT transformation and ShuffleNet deep neural network.The specific content is as follows:1.Study a medical image zero-water sign algorithm based on the LPT-DCT algorithm.First,use the LPT-DCT combination to transform to extract the characteristic vector of medical images,and then select the 32-bit low-frequency coefficient of the feature zone and combine to perceive the hash algorithm to generate 32-bit feature vectors.Use the Logistic Map chaos technology to disrupt the original watermark information to generate a two-value sequence of encrypted encryption,and then use zero-watermark technology to achieve watermark embedding and extraction process.The results confirmed that the watermark extracted by the algorithm has a good robustness and can effectively resist conventional attacks and rotation attacks.2.Study a zero-watermark algorithm based on the ShuffleNet pre-training model.The basic idea is using the ShuffleNet pre-training model to extract medical image feature vectors.Use image feature vectors to realize the embedding and extraction of zero watermarks.Enter the medical image to the pre-trained ShuffleNet network,take the output data of the full connection layer,and then perceive the Hash through DCT to obtain the characteristic vector of the medical image.Use the feature vector of the image and the zero watermark technology to realize the zero watermark of the medical image.Embedded and extracted.3.Study a medical image robust watermark algorithm based on the ShuffleNet network migration.Specifically,the ShuffleNet network after migration is used to extract the characteristics of medical images,and uses feature vectors,heterogeneous or logic and zero watermark concepts to realize the embedded and extraction of the zero watermark of medical images.The impact of different conventional attacks and geometric attacks on the digital water marks algorithm,and evaluated its robustness,hoping to further improve the robustness of digital watermarks.Experiments have proved that the algorithm has good anti-geometric attack capabilities.