Study On Digital Watermarking For Document Security

With the fast development of information-networking and printing-technology, reproduction and dissemination of the documents are becoming very easy, so the document security is facing extremely serious challenges. Digital watermarking, as a potential resoluble approach, receives much attention. This thesis focuses on enhancing the robustness of the document image watermarking against geometric attacks and print-scan operations, and proposes document image normalization options in order to increase the performance of watermarking against geometric attacks. The purpose is to present more practical watermarking algorithms and systems for document security.As a part of applied research, the watermarking system solutions are particularly important. This thesis presents the framework of document security systems based on watermarking algorithm in binary document images for the document security during transmission. The framework includes the document image normalization, the embedder and detector of binary document watermarking.In order to solve the image noise, tilt, zoom and other issues during the print-scan operation and the transmission of the document image, this thesis presents a new image normalization method, including images binarization, noise removal, skew detection, and character segmentation. The normalized image meets a set of predefined criteria with respect to the angle of skew, the size of word and the distance from text, and the adopted affine matrix is obtained through skew detection and character segmentation. The experimental results show that the operations of each step of the method have reached the desired results, which ensure the validity of the watermark embedding and the accuracy of detection and extraction, thus increase the performance of the document image watermarking against geometric attacks.Document security requires the robustness for geometric attack and PS operation, however, the difficulty of watermarking increased largely because of the characteristics of the document image. To solve this problem, we propose a novel double domain watermarking, which uses transform domain signal to enhance the robustness of the watermark. The watermark pattern is generated as a transform domain signal, and adapted to the characteristics of human visual system by adjusting the localized embedding strength in the spatial domain. It allows us to use the features of double domains for embedding, and avoids transforming the original binary image directly, which introduces unacceptable quality degradation. The theoretical analysis and experimental results show that the presented scheme manifests the good robustness against common image manipulations, geometrical attacks as well as print-scan process.