Research On Some Key Issues For Digital Image Watermarking In The Transform Domain

Digital watermarking as one of techniques can be utilized in copyright protection and content authentication. Since 1990s, the increasing requirement of multimedia forensics encourages the development of digital watermarking. However, digital watermarking is still in its infancy and remains a great deal of problems. Therefore, this dissertation takes still images as research object, and presents several image watermarking methods by incorporating the advantages of the features of image, visual model, etc,and provides some effective solution for watermarking in practical application. Contributions of this thesis can be summarized as following:(1)A hybrid approach of DWT (discrete wavelet transform) and DCT (discrete cosine transform) for the rational dither modulation (RDM) watermarking is proposed in this thesis. Considering some sensitivity problems of conventional quantization-based watermarking against amplitude scaling, JPEG compression attack, etc, we propose an adaptive embedding strategy with RDM by selecting the richest texture region as the embedding space. The synchronization of the quantization step-size can be adaptively controled by using this scheme. Experimental results confirm the imperceptibility of the proposed watermarking and its higher robustness against attacks compared to alternative watermarking methods in the literature.(2)A robust quantization-based watermarking is proposed in the dual tree complex wavelet transform (DTCWT) domain. We propose an embedding method by taking advantages of the dual tree complex wavelets (perfect reconstruction, approximate shift invariance, and directional selectivity), and we obtain the optimum of scalar factor by concerning the relation among the scalar factor, distortion and robustness. Therefore the trade-off between the imperceptibility and the robustness can be achieved effectively. Experimental results demonstrate that the proposed method is robust against JPEG compression, additive white Gaussian noise (AWGN), and some kinds of geometric attacks such as amplitude scaling, rotation, etc. Furthermore, on the basis of the quantization-based watermarking, a feature-based geometric invariant is constructed by combining the image normalization technology with the scale invariant feature transform, and an image blind watermarking method is presented based on key entropy in the DTCWT domain. Using the geometric invariant in the proposed method, the sensitivity problem can be solved for the watermark when against common geometric attacks. Experimental results show that it has good robustness against some kinds of geometric distortions consist of rotation, scaling, and affine transformations, etc.(3)A blind image watermarking is proposed in the wavelet domain by using dual embedding scheme. According to the disadvantages of watermarking which the watermark is embedded only in low-frequency region or in high-frequency, two embedding factors are exploited by considering the visual perceptual characteristics, and they are used in the low-frequency and high-frequency domain for watermark embedding respectively. Thanks to the good multi-resolution of discrete wavelet transform and the similarity with human vision perceptual model, the image quality can be guaranteed and the robustness can be improved against common image processing and geometric attacks by using the proposed approach.(4)A multiplicative watermarking based on visual model is proposed. A multiobjective optimization model is exploited by taking the imperceptibility and the robustness as object functions. Using this scheme, the conflict between the imperceptibility and the robustness can be sloved effectively. Further, we investigated combining the approximate shift invariance of DTCWT with energy scheme. Thus the robustness of watermarking can be improved. Experimental results show that the imperceptibility of the proposed is good, and the watermark can be robust against common combinational attacks.(5)We give an insight of some DTCWT-based image processing, and then propose an image diffusion filtering algorithm based on dual tree complex wavelet and wave atoms. Under the background of anisotropic diffusion model, we introduce a novel computational technique for diffusivity by taking advantages of the properties of the wave atoms (good sparse representation) and the dual tree complex wavelet transform(perfect reconstruction, approximate shift invariance and directional selectivity). Using this technique, the direction and strength of image diffusion can be controled effectively, and the problem which many image detail features are wiped away easily by applying the traditional anisotropic diffusion model can be solved. Experimental results show that many features of image such as edges and textures can be preserved well after filtering via the proposed algorithm. Finally, we introduce an adaptive anisotropic diffusion image filtering by considering the correlation and localization among the complex wavelet coefficients, and we propose an image denoising method based on PDEs (Partial Differential Equations) in the DTCWT domain under the theory framework of the proposed algorithm. Then an invented patent has been holding on this achievement, and the patent No. is 201010107623.0.