Research On Some Key Algorithms Of Audio Digital Watermarking

In recent years, the rapid development of Internet technology provides a great convenience for the dissemination of digital audio content, but also result in copyright infringement and malicious tampering. How to provide management and protection of copyright and integrity for digital audio content has become a research hotspot in the academic circles. Traditional encryption and digital watermarking techniques are both used in the copyright protection and integrity protection of digital audio content to solve the problem of copyright infringement and malicious tampering. However, since the problem of the encryption technology itself, it is not applicable to the copyright protection and integrity protection of digital audio content. Audio watermarking, due to its robustness and imperceptibility, is causing increasing concern in the academic circle and it is an effective way to protect the copyright and integrity of digital audio content. Audio watermarking protects the copyright and integrity of digital audio content by embedding audio watermarking sequence into the digital audio contents in an imperceptible way. Current research on audio watermarking focused on how to design the audio watermarking algorithm to make sure it has superior robustness, imperceptibility and high embedding capacity.This thesis studies some key algorithms of audio watermarking based on the copyright protection and integrity protection of digital audio content. During the design process of the proposed algorithms, this thesis references the latest academic achievements in other fields, such as the Empirical Mode Decomposition, Compressive sensing and machine learning, to improve the tranditional watermarking algorithm. In detail, main achivements of this thesis are summarized as follows:1. An audio watermarking algorithm based on empirical mode decomposition and particle swarm optimization is proposed in this thesis. Residual signal components decomposed from EMD were selected to embed watermark by the proposed algorithm and the watermarked components are proved to be still residual signal components, which provides the theoretical basis for the watermark embedding and extraction. The optimal audio watermark embedding strength in EMD was solved by particle swarm optimization algorithm. The robustness and imperceptibility of watermarking can be satisfied according to this embedding strength. The simulation results show that using the optimal embedding strength to embed watermark can guarantee the imperceptibility and robustness of the watermark in most attack conditions.2. A robust audio watermarking algorithm which has high imperceptibility is proposed in this thesis. The intrinsic feature of the final residual decomposed from the audio frame is selected to embed watermark and the algorithm works by shifting each element of the final residual to make its sum greater or less than 0. The experimental results show that the proposed algorithm does not change the property of the final residual after embedding watermark and the watermark is robust against various kinds of attacks. Compared with existing classic algorithms based on EMD, the proposed algorithm largely improves both the robustness and imperceptibility of watermarking.3. A novel compressive sensing-based audio semi-fragile zero-watermarking algorithm is proposed in this thesis. This algorithm transforms the original audio signal into the wavelet domain and applies compressive sensing theory to the approximation wavelet coefficients. The zero-watermarking is constructed according to the positive and negative properties of elements in the measurement vector. The experimental results show that the proposed algorithm is not only robust against common audio signal processing but also fragile to malicious tampering. Compared with the existing algorithms, the proposed algorithm improves malicious tampering detection accuracy in common audio signal processing environments.4. An audio watermarking mechanism based on variational bayesian learning is proposed in this thesis. The spread spectrum watermarking was embedded in the DCT coefficients of audio frames. MFCC features extracted from watermarked audio frames as well as un-watermarked ones were trained to establish their Gaussian mixture models and to estimate the parameters by vatiational Bayesian learning method respectively. The watermarking was detected according to the maximum likelihood principle. The experimental results show that the proposed algorithm can lower the false detection rate compared with the method using EM algorithm when the audio signal was under noise and malicious attacks. Also, the experiments show that the proposed algorithm achieves better performance in handling insufficient training data as well as getting rid of over-fitting problem.The expetimental results and performance analysis show that, the work done in this thesis improves the performance of existing similar algorithms and has important theoretical and practical significance to improve the research on audio watermarking algorithm and solve the problem of copyright Infrigement and malicious tampering of digital audio content.