Research On Digital Watermarking Technology Based On Self Supervision

The development of modern computer technology and the popularity of multimedia technology have brought convenience to the sharing of multimedia information,but also exacerbated the occurrence of digital piracy.Digital watermarking technology is an information technology that embeds copyright information into multimedia images in an implicit way,which can trace copyright infringement and effectively protect copyright.This article focuses on improving the comprehensive performance of digital watermarking technology in robustness and imperceptibility on widely used static images,and explores some innovative methods.The main work and innovation points are described as follows:A self-supervised digital watermarking method is proposed,which improves the convolutional neural network structure.The model consists of a self-supervised encoder and decoder.Firstly,the encoder and decoder are trained with watermark-free images,where the self-supervised encoder compresses the image into a fixed-length feature vector,and the decoder decodes the vector to the original image,preserving both the texture and deep features of the image.Then,the self-supervised encoder is retrained with watermarked images to fuse the watermark information into the original image and obtain a watermarked image.During watermark extraction,the network takes the latent vector of the original image as input,and outputs the latent vectors of both the original and watermarked images,computing their difference.By comparing the extracted watermark information with the original watermark information,any tampering of the watermark can be detected.Our proposed method achieves a certain degree of improvement in both the imperceptibility and robustness of the watermarking.A self-supervised zero-watermarking method is proposed,which takes static images as carriers and fully considers possible attack methods to improve the robustness and reliability of watermarks.By embedding specific marks into the image,the copyright and content security of the image can be effectively protected without affecting the quality and information integrity of the image.This method vectorizes the original image based on the previous method,extracts the features of the image,and embeds the feature mark into the image.The loss is calculated and the Adam optimizer is used to optimize the image,and the final output contains the zero watermark image.The decoder detects the zero watermark image,extracts the mark,and compares it with the original mark to verify the accuracy and completeness of the zero watermark.In addition,different types of attacks(such as noise attacks and adversarial attacks)are introduced during network training to improve the robustness of the image against noise attacks.Experimental results show that the proposed method has higher robustness and reliability compared to traditional watermarking technology.This method can maintain good watermark recognition and extraction performance under different types of attacks,such as rotation,scaling,cropping,and noise attacks.Moreover,the method has a high watermark embedding and extraction speed and is suitable for various practical application scenarios.