Image Digital Watermarking Techniques Based On Coefficient Characters Of BOR Multi-wavelet

It is known to us that "Science and Technology is a double-edged sword", the digital information revolution has brought a lot of convenience to people's lives, while at the same time, it gave rise to difficulties. Urgent need for digital product protection and information security greatly promoted the research and development of digital watermarking -- an important branch of information hiding technology, in which the multimedia watermarking technology based on wavelet transform is a relatively active research area in recent years, especially with the wavelet transform applied to JPEG2000 standard, the research in this field becomes even more practically significant. Multi-wavelet is a natural promotion of a single wavelet, who has characters as orthogonality, symmetry, short support and high vanishing moments at the same time, which are impossible for a single wavelet. In the practical application of image processing, it is very important for an analysis tool to possess these two properties: both orthogonality (to maintain energy) and symmetry (linear phase). Through the active explorations, people have made certain achievements in image processing applications, such as still image compression, coding, denoising and etc; however, the study on information hiding in image is fewer. Lack of understanding of the characteristics of multi-wavelet transform makes watermarking algorithm based on multi-wavelet domain lost its vigor and vitality. Therefore, information hiding techniques based on self-structural features of multi-wavelet transform yet still needs to be further studied. Under this practical background, a thorough research on image digital watermarking techniques based on coefficient characters of BOR multi-wavelet is given. The main innovative contents in this dissertation are listed as following:Firstly, on base of detailed analysis of the decomposed coefficients of BOR multi-wavelet, the"layer"concept of an image in multi-wavelet domain is proposed, that is, by multi-wavelet transformation, an image is decomposed into four layers, in a sense, the sub-image of each layer is equivalent to the one in scalar wavelet domain, which has its own low-frequency parts and high-frequency parts. Each layer sub-image is four interposed by rows and columns respectively and convoluted by its own sets of reconstructive filters, then the sum of these four reconstructive sub-images is the original one. For this unique character, data can be hiden in single layer sub-image or two layers, three layers,even four layers simultaneously, thus the selection of hiding layer undoubtedly gives another secret key and this concept also shows us the flexibility of hiding data in multi-wavelet domain.Secondly, an adaptive blind watermarking algorithm is put forward to embed a binary image into the first layer sub-image and four layers at the same time in BOR domain respectively. For the former, the watermark shows up excellent robustness which indicates the potential advantages of information hiding capacity in multi-wavelet domain; and for the latter, the watermark image takes on the perfect reconstruction quality of BOR-base, and the watermark has strong resistance to compression. Experimental results show that compared with the algorithm in Haar domain, the above two options in multi-wavelet domain prevail on.Thirdly, multi-wavelet transform along with the traditional scalar wavelet (single wavelet) transform are both multi-scale decomposition, but the obvious difference between them is that for multi-wavelet, according to"layer"perspective: in the same direction on the same scale, there exist four layer sub-images. This feature of multi-wavelet was showed evidently by the sub-images (low-frequency coefficients) of coarse-scale which focus the majority energy of the original image. The concept of average sub-image of the coarse (low-frequency) sub-images in DMWD is put forward. And at the same time we demonstrate that the above average sub-image is more robust than any other sub-images either in DWD or in DMWD under the attacks of different image processing operators.Fourthly, in theory, watermarking algorithms based on low-frequency coefficients of wavelet-domain are equally applicable to the low-frequency coefficients of multi-wavelet domain and can get a better robustness. From this point of view, firstly, we improve the Hong algorithm in literature [133] to embed watermark into the low-frequency coefficients in Haar domain (abbreviated Hong low-frequency algorithm), under the non-perceptual premise, robustness of the watermark improved significantly. Secondly, to further improve the Hong low-frequency algorithm we embed the watermark into the above average sub-image of low-frequency coefficients in BOR domain. Theory and simulation experiments both confirm the robustness of the algorithm, especially, its strong resistance to compression and scaling is better than other existing algorithms, while the algorithm can effectively resist printing and scanning attacks that would fail a variety of algorithms. So it is recommended that the algorithm be used in practice.At last, a simple while robust zero-watermarking approach is suggested based on the double-average method, in which the watermarking bits are extracted from the average sub-image by comparing the center coefficient value with the mean of its four neighbors.There are six chapters altogether in this dissertation.In chapter one, background of this research, the fundamental framework and the relative issues of the watermarking techniques in discrete wavelet (including scalar wavelet and multi-wavelet) domain are sketched, and the present study status at home and abroad is presented completely.In chapter two, the basic theory of multi-wavelet is detailed on, including the definition of multi-wavelet, mathematical properties of multi-wavelet and its detailed image analysis steps for digital images, which are mathematical basis behind this study. End of this chapter, the protagonist -- BOR Multi-wavelet and its traits applicable to watermarking technology are introduced, it is these advantages that make it the preferred multi-wavelet base in watermarking field.In chapter three, the"layer"concept (different from the decomposed levels) of an image in DMWD is proposed, that is, by multi-wavelet transformation, an image is decomposed into four layers, the sub-image of each layer is equivalent to the one in DWD (discrete scalar wavelet domain), which has its own low-frequency parts and high-frequency parts. For this unique character, data can be hidden in different layer sub-images, the selection of hiding layer undoubtedly gives another secret key. An adaptive blind watermarking algorithm is put forward to embed a binary image only into the first layer sub-image according to above trait. Rigorous analysis and experimental results show that compared with traditional embedding method in DMWD, the given technique can reduce the computational complexity, increase the computing speed, and the watermarked image shows up the good imperceptibility and high robustness.In chapter four, the concept of average sub-image of the coarse (low-frequency) sub-images in DMWD is put forward. And at the same time we demonstrate that the above average sub-image is more robust than any other sub-images either in DWD or in DMWD under the attacks of different image processing operators. A new watermarking algorithm based on the above sub-image is raised which can stand against lots of attacks including the fatal enemy: printing and scanning who ever failed many algorithms in watermarking fields. In fifth chapter, we detail on the proposed simple while robust zero-watermarking algorithm in BOR domain and give the experimental results.And finally, contents of the whole dissertation are summarized and the future research direction is predicted.