Research On Quantization-based Watermarking With Robust Visual Model

With the rapid development of digital multimedia technology and the rapid spread of low-cost digital equipments, a variety of Internet multimedia information service has been widely used. The advanced information technology has brought convenience for us. However, illegal reproduction and convenient manipulation of digital media cause considerable economic losses to the media creators and the con-tent providers. As an effective technology, digital watermarking has been widely studied as an essential tool for multimedia copyright protection and data authen-tication. Among these watermarking schemes presented so far, quantization-based watermarking schemes can embed the watermark bits by quantizing the host signal samples with a set of quantizers. It has received widespread attention and provides a computational efficient watermarking algorithms with high data capacity.In the quantization-based watermarking framework, human visual system (HVS) has a great impact in the watermark embedding. With the development of modern optics and optoelectronic imaging technology, lots of researches have been conducted over the years to understand the HVS to obtain a better tradeoff between fidelity and robustness. Unfortunately, the HVS is image dependent and can rely on adapt-ing to local image properties. Therefore, the perceptual characters calculated in the original image are inconsistent with those in the watermarked image. So their applications are still limited in the quantization-based watermarking framework.Focusing on the issues existing in the quantization-based watermarking frame-work, robust HVS and the related watermarking schemes are studied in this dis-sertation. The traditional visual model is improved and the Weber law is applied into the logarithmic STDM watermarking scheme. Various prediction modes are used to improve the robustness for reference block selection, which can provide a significant improvement on the watermarked image fidelity. Based on the Weber’s law, the watermarking scheme can feature more perceptual advantages, and the weakness due to collusion attack can be overcome by the uniform quantization in the logarithmic domain. Moreover, the JND modeling has been studied because the good performance can be achieved by introducing it into the watermarking. A new robust visual JND model and visual attention-based JND model are proposed and the related logarithmic spread transform dither modulation (STDM) watermarking schemes are proposed to improve the robustness.The main contributions of the dissertation are as follows:1. For the vulnerability of the reference block selection in the quantization-based perceptual watermarking algorithm, an prediction modes based STDM wa-termarking scheme is proposed. As the watermarked pixels and the pixels in the current block has a strong similarity, the predicted block is more close to the current block for Euclidean distance. The fixed prediction ensure the robustness of the reference block selection.2. Based on perceptual model, an logarithmic spread-transform dither modula-tion method is proposed, which incorporates a novel logarithmic function into the STDM framework with the improved quantization step. In order to avoid complicated calculating process in the original logarithmic quantization-based method, and to provide invariance to the amplitude scaling attack, an im-proved transformed function is selected for the proposed watermarking scheme. Furthermore, the perceptual model is further exploited to adjust the quanti-zation step adaptively for watermark embedding.3. A new perceptual JND model, which gives us a novel way to model the HVS in the watermarking algorithms robustly and accurately, is proposed. In this regard, the new measurement of the pixel intensity and the edge strength is introduced to calculate the slacks at the watermark embedder and detector, despite the modifications to the DCT coefficients introduced due to the wa-termark embedding. Furthermore, an improved logarithmic STDM scheme is further proposed based on this model. Experiments are conducted to show the validity and robustness of the proposed perceptual JND model and the presented watermarking scheme with the robust perceptual JND model can provide a more satisfactory robustness performance.4. The existing JND model treats every region in the image with an equal at-tention level. Visual attention, which reflects the visual attention, is proposed to modulate the perceptual JND model. Based on the improved JND mod- el, a logarithmic spread transform dither modulation (STDM) watermarking scheme is proposed. Simulations show that the proposed watermarking scheme with the improved JND model has superior robustness compared with existing STDM schemes...