Research On Digital Watermarking Techniques For Image Based On Multiwavelet Domain

This thesis is concerned with one of the most popular research topics in the fieldof information hiding-- the digital watermarking technology, with emphasis placed onits applications in digital images. Specially, this thesis focuses on the research andimplementations of watermarking techniques based on multiwavelet for copyrightprotection of digital images. Firstly, the states, basic model, characteristics, classification and applications ofthe digital watermarking techniques are introduced. Strategies of currentwatermarking techniques are broadly investigated. The current methods of the design,generation and pretreatment of watermark are listed out. The various attack methodsare introduced and fully analyzed and several directions of development of digitalwatermarking at next stage are proposed. Secondly, we introduce the relevant theoryof multiwavelet decomposition of images and analyze characteristics of multiwaveletcoefficients. Like a scalar wavelet case, the multiwavelet decomposition provides us amultiresolution representation of an image. However, there are four differentcoefficient subblocks in the same direction and in the same resolution. In addition to,a multiwavelet has some advantages in comparison to a scalar wavelet, such as shortsupport, orthogonality, and symmetry. Then, the multiwavelet theory, the HVS(human visual system) theory and Singular value decomposition(SVD) are applied todigital watermarking embedment. According to the characteristics of the original image, two adaptive embeddingstrategies based on HVS are proposed. One is based on the feature of visual maskingof human vision system. Using the tree structure relation of wavelet coefficients, thefirst component coefficients in the coarse level are classified into two groups.According to the results of classification, watermark components with differentstrength are embedded into these wavelet coefficients. Another is based on theinter-block correlation of DMWT coefficients. After multiwavelet transformation ofan image, the wavelet coefficients are formed into wavelet tree. According toneighborhood averaging of the wavelet tree group and the local feature of an image, abinary pseudo-random watermark components are repeatedly embedded into waveletcoefficients. The watermark can be extracted without the original image. For the stabilityand invariabilityof rotation of singular value, a novel embeddedstrategy in dual transform domains is proposed. It makes full use of themulti-resolution of DMWT and the characteristic of SVD and embeds a binary imagewatermark into the original image by quantification process. The watermark can beextracted without the original image, during which those who don't know the secretkey can't rightly retrieve the watermark. The algorithms above are implemented. Some performance analyses comparedwith other relevant algorithm are drawn out. The experimental results demonstratethat the present algorithms are of robustness and good imperceptibility.