Research On Blind Watermarking Technique Of Digital Image

With the rapid development of computer and network technology, informationsecurity and copyright protection of the digital media have become an urgent problem.As an important means to protect the digital media effectively, the digital watermarkingtechnique has become a hot research topic. In this thesis, the background, classification,application and recent development of the digital watermarking are given first. Then theuniversal model of the technique is set up, the design requirements and their correlationare analyzed, and the method to evaluate a watermarking system is provided. In addition,a new classification for different watermarking algorithms is proposed. Taking stillimages as example, the thesis utilizes several techniques and methods such as chaos,wavelet transform, image scrambling, neural network, invariant moments andhypothesis testing etc. The work is focused on the blind watermarking technique in thespatial and wavelet domain. The contributions of this thesis include:Because the chaotic map has fine characteristics and few researches are focused onthe 2-D chaotic map, two kinds of 2-D chaotic maps and a 1-D chaotic map aredesigned. Their statistical characteristics are analyzed. They are applied to the securityof the watermark methods in three different ways to make the watermark security. Threemethods of blind watermarking methods based on chaotic maps in wavelet domain arepresented.(1) A 2-D shift chaotic map is designed. The embedding addresses of the watermarkin the wavelet domain of the image are determined by the chaotic sequence, and therepeated addresses are avoided by conflict processing. On the other hand, the watermarkis scrambled before embedding for security. Finally, every watermark bit is embeddedrepeatedly in the trees in wavelet domain, so the robustness is enhanced.(2) A 2-D congruent chaotic map is designed to scramble and expand thewatermark, and the security is improved. Then several pairs of coefficient blocks in thewavelet domain are selected randomly, and the relations between the two summations ofeach two blocks are remained or changed according to the watermark amplitudedetermined by the noise visible function. So the watermark embedding is finished in thewavelet domain. Finally, a new method using F norm to implement similaritymeasurement of the extracted watermark and the original watermark is presented.(3) An improved 1-D shift chaotic map is designed, and the binary chaotic sequenceis used to modulate the watermark. Then the feature points of the lower frequencysubband are extracted using the improved Achard-Rouquet operator, so the watermarkinformation is produced in virtue of the feature trees composed by the feature points.The watermark is stable. The method is a zero-watermark method which has nodistortion and is adaptive to the image. In addition, the method is simple and secure, andit can be applied to the multi-watermark technique.The geometrical attacks on the watermark are discussed, and two blindwatermarking methods which resist geometrical attacks such as rotation, scaling andtranslation are proposed.(1) Firstly, the feature points of the image are extracted using the improved Harrisoperator, and the Delaunay triangle lattice is produced according to the triangle growingmethod. Secondly, the watermark information is constructed utilizing the geometricalinvariant characteristics in the triangles. Finally, to ensure the security of the watermark,the logistic chaotic sequence is introduced to accomplish chaos search, so the key isgenerated. The method is also a zero-watermark method.(2) Making use of combined invariants, Hu's invariant moments and neuralnetwork, the relation between the watermarked image and the watermark is established.When detecting the watermark, the distance of the invariant moments feature isproposed to calculate the similarity. The method can resist some attacks such as rotation,translation, scaling and blur etc.The blind extraction model of the meaningless watermark is studied. It is abit-by-bit extraction method based on the hypothesis testing, and it can extract every bitof the watermark other than the common methods which just judge whether thewatermark exists or not. The model is realized in the spatial and transform domain. Inthe spatial domain, based on the χ2 goodness-of-fit test, a gauss mixture model isestablished. The watermark amplitude is determined by the local entropy and the imagegray. In the wavelet domain, a general gauss distribution is established for the waveletcoefficients. The watermark amplitude is determined by the standard deviations of thelocal characteristics which are estimated by the context modeling. The watermark is thebinary chaotic sequence produced by two level chaotic maps. It has goodautocorrelation, cross correlation, balance and security. Finally, an improved method ispresented.