Research On Network Method For Text Image Watermarking

With the innovation and development of internet information technology,the transmission and circulation of text content have become increasingly convenient,making text information an important component of online communication.Information technology not only promotes the dissemination of text content,but also brings the need to protect text copyright.Therefore,digital watermarking technology for text content has become an important research direction.In recent years,research on text image watermarking technology has emerged one after another,mainly divided into spatial domain based methods and deep learning based methods.However,existing methods have drawbacks such as inability to apply to different types of fonts and insufficient robustness.In this paper,the text image watermarking method based on deep learning is studied by combining attribute features and Semantic information.The innovative achievements achieved are as follows:1.The watermarking network method for single character text images is proposedThis dissertation proposes using common font attributes such as stroke thickness and edge sharpness to express watermarks,which are suitable for various fonts;And introduce a noise layer into the network to enhance the robustness of the watermark.Specifically,this dissertation uses a residual block structure to construct a convolutional neural network to extract the character attribute feature vectors of the carrier image,combines the watermark information to generate a watermark image,and adds a noise layer to the network to simulate distortion in actual scenes,enhancing the robustness of the watermark network.The experimental results show that the single character text image watermarking network method proposed in this dissertation has higher watermark extraction accuracy and visual quality compared to existing text watermarking algorithms,while also being more robust and suitable for various fonts.2.The watermarking network method for multi character text images is proposedAiming at the problem that single character text image watermarking method may face the difficulty of character segmentation in practical application,this dissertation proposes a watermarking network directly applicable to multi character text images based on Semantic information.In this dissertation,the Semantic information extraction module is used to obtain the semantic features of the carrier image,and the convolutional neural network is used to extract the image features of the carrier image.In addition,in order to more realistically simulate real-world distortion changes,a richer variety of change functions have been added to the noise layer of the network,further enhancing the robustness of the watermark network.The experimental results show that the multi character text image watermarking network method proposed in this dissertation can cope with the difficulty of character segmentation and has better robustness than existing text image watermarking methods.