Research On Some Key Technologies Of Digital Audio Watermarking For Copyright Protection

With the rapid development of Internet and multimedia technology,people can more convenient acquire the multimedia resources what are needed.As the important part of multimedia resources,digital audio is becoming easier to be illegally tampered and transmitted,which makes the problem of digital audio security becoming very serious.The digital audio watermarking has been proposed as one of the most effective technology of protecting the legitimate rights of the audio copyright owners,in which providing a means to embed a unique code as a "fingerprint" into each copy of the distributed digital audio.In recent years,great progress has been made in the research of audio watermarking algorithms.However,the existing algorithms are difficult to resist de-synchronization attacks and achieve good balance between the robustness and imperceptibility.In this paper,we propose three effective digital audio watermarking algorithms to deal with these problems,which can be summarized as follows:1.Combining the Cosine Modulated Gaussian(CM-Gaussian)filter and the stability of low-frequency coefficients of 1-D stationary wavelet transform,we propose a robust feature point based on wavelet domain audio watermarking algorithm.Firstly,the lowpass subband is obtained by using 1-D stationary wavelet transform,and the response gradient of each coefficient is calculated by using 1-D CM-Gaussian filter.Then,the robust feature points with stable and uniform distribution are detected,and the invariant local regions are built adaptively according to local audio content.Finally,the watermark is embedded into local region of each feature point using adaptive quantification method.Experiments show that the proposed algorithm not only has good imperceptibility,but also has better robustness to common attacks and de-synchronization attacks(especially local de-synchronization transformations).2.Using the good ability of the truncated generalized Cauchy to describe the distribution of transform domain coefficients,a new blind detector is proposed for the SWT-based multiplicative audio watermarking,wherein a PDF based on the TG Cauchy distribution is used.Firstly,the watermark data is embedded into the significant high-frequency coefficients in SWT domain.Secondly,a blind watermark extraction approach is developed using the maximum likelihood decision rule.Simulation experiment results show the proposed algorithm provides an excellent watermark invisibility and outstanding robustness,which can be resilient to various kinds of attacks.3.Based on fully considering the correlation of 1-D transform domain within the subband coefficients and with the intra-scales,we propose a dual-tree complex wavelet domain audio watermarking algorithm by modeling multivariate BKF distribution.Firstly,the audio is decomposed into two highpass subbands by applying the 2-level 1-D dual-tree complex wavelet transform,and the important coefficients of the two highpass subbands are modified using the multiplicative method.Then,the multivariate BKF distribution is utilized to model the two highpass subband coefficients by considering the correlations of the coefficient between the subbands at different scales,and the shape parameters of the model are estimated by the coefficients of 1-level highpass subband which are not embedded watermark.Finally,the specific watermark information is blindly extracted using the maximum likelihood decision rule.Experimental results demonstrate that the proposed approach can provide a better imperceptibility and robustness in comparison with the recently proposed techniques.