Nonlinear Detector Text Image Watermarking Research

With the rapid development of the science and technology, the usage of the Internet is very popular. Many kinds of digital multimedia information spread in the network. The digital information in communication is easy to be edited or copied, even being maliciously damaged. Owing to the development of the related theories and technique, the digital watermarking emerges. Then, the digital watermarking has been widely adopted for protecting information in many fields. For instance, the digital watermarking is frequently embedded into the color or gray images. The text images, such as the public document, the academic certificate and the medical record, are also spread over and stored in the internet, which needs to be protected for the information security. Therefore, the studies of the digital watermarking for text images are of practical interest.This thesis mainly investigates the watermarking scheme for the text images, and proposes a nonlinear detector. First, the digital text documents can be regards the images. Then, the discrete cosine transformation(DCT)coefficients of the text image are reordered by certain rules. Then, from the reordered DCT coefficients, we select a segment to be embedded by the watermarking signal. For the reordered DCT coefficients by their absolute values, we assume their probability density function (PDF) accords with that of the dichotomous noise. Then, under the assumptions of weak watermarking signal and the proposed PDF model, we develop a nonlinear sign detector to detect the digital watermarking signal in text images by the Neyman-Pearson theorem. Furthermore, for sufficiently length of watermarking signals, the statistical characteristics of this nonlinear detector are analyzed in detail. Using the asymptotic distribution of the test statistics, the response threshold of the nonlinear detector is theoretically given. Extensive experimental results demonstrate the robustness of watermark against some common attacks, e.g., JPEG compression, cropping, filtering, additive Gaussian noise, dithering, and also verify the robust performance of the nonlinear sign detector for watermark detection. Also, we compared the classical watermarking method of Barni with the nonlinear watermarking algorithm by the measure of reflecting coefficient. The experimental results show that the performance of the nonlinear approach is better. However, the proposed watermarking detection process requires knowing not only the watermark, but also all marked DCT coefficients positions. According to the abovementioned experiments, its robustness capability against noisy corrupting and cropping are outstanding, but needs to be further improved to be against the desynchronization attack.