Research On The Key Technologies Of Digital Watermarking

Digital watermarking has provided an effective method for copyright protection and has gained great popularity in multi-media security domain in recent years. In this paper the key technologies of digital watermarking including watermarking preprocessing, multi-watermarking, detection theories and watermarking based on intellective computation are researched. Main contents of this dissertation are as follows.Firstly, watermarking preprocessing is researched. An equitable matrix's elements are correlated and its equitable state is stable in some degree, so a watermarking algorithm including an equitable matrix preprocessing is proposed. The preprocessing is at first to scramble the watermarking information, then convert the scrambled information to corresponding equitable matrix and at last the equitable matrix is embedded into the host image. Through preprocessing the embedded watermark can be extracted at two levels that are private and public respectively.Secondly, multi-watermarking is researched. A multi-watermark iterative blending algorithm is given based on present single watermark iterative blending algorithm. Weak robustness of the iterative blending algorithm is proved by theory analysis and experiments. Main reason of weak robustness is that the watermarked data's error is multiplied at many times so the embedded watermark can't be extracted rightly. In order to enhance robustness, the watermarked sub images are modified by the nearby sub images, at the same time several watermark images are embedded at different places decided by secret keys. The second part of the chapter is about morphological multi-watermarking. According to the properties of morphological operations on images, a morphological multi-watermarking scheme, in which the morphological modes and a visible watermarking are embedded in the host image, is proposed. If every mode corresponds with a future user, digital products for different users can be made one time. The embedded visible watermarking requires future users should remove it and in the process digital product is signed. So a typical copy tracing application is realized.Thirdly, non-linear correlation-based detectors and sequential watermark detector are researched. The traditional correlation-based detector is optimal only for the Gaussian data, but the Laplacian PDF is more appropriate to model the coefficients in discrete ridgelet transform (DRT) domain. An additive maximum-likelihood detector based on the Laplacian PDF is analyzed and theoretical result of its performance is given. The experiments show that error of the Laplacian model for DRT coefficients is smaller than that of the Gaussian model. About the second part, fixed sample size (FSS) watermark detection needs a large number of signal observations and it is not suitable for applications such as detecting multi-watermarking or video watermarking detection. The sequential watermark detection allows simultaneous monitoring through increasing signal samples, so it can overcome the shortcomings of FSS watermark detection. In analysis of sequential watermark detection, we find that operating characteristic function(OCF) and average sample number(ASN) are both related with the actual embedding factor. In order to improve the sequential watermark detector performance, a local network is applied to predict the original image because it can reduce prediction error compared with the simple neighboring pix prediction.At last, a vector quantization watermarking algorithm based on an ant colony system(ACS) is proposed. Before optimizing watermarking system, ant colony system is improved. According to stage properties of bionic optimization algorithms, an improved ant colony algorithm DSACA is proposed to enhance speed and accuracy of ACS. The process of watermarking algorithm is as follows. At first, error model of vector quantization, watermarking embedding and noisy channel are established and further the simplified model is given according to real circumstances. Error of the watermark embedding and the noisy channel are directly related to the codeword index order, at the same time deciding the codeword index is a typical NP-hard problem, so improved ACS is used to optimize vector quantization watermarking algorithm to make bit error rate lower and embedded watermark more invisible.