The Study On Key Technologies In Digital Audio Watermarking

In recent years,the fast development of digital audio processing technology and the popularization of Internet make it dead easy for people to access,edit and distribute digital audio works.The copyright protection of digital audio works becomes a burning question.As a potential solution,digital audio watermarking technology is paid extensive attention by researchers at home and abroad and becomes an international academic focus.In this paper,an intensive study is made of the key technologies,including zero watermarking,multiple watermarking and multipurpose watermarking,in digital audio watermarking system.The main contributions of this dissertation are as follows:(1) Since the human auditory system is extremely sensitive,it is very hard to achieve the balance between the robustness and transparency in audio watermarking system.To solve this difficult problem in theory,a study is made of the audio zero watermarking technique.First,in view of the transient steady property of natural speech signal,a speech zero watermarking algorithm based on the vector quantization of linear prediction coefficients is proposed.This algorithm achieves good detection reliability and robustness.Then,an audio zero watermarking scheme based on the combination of the multi-resolution analysis property of discrete wavelet transform,the energy compression property of discrete cosine transform and the Gaussian noise suppression property of higher-order cumulant is proposed.This algorithm has great robustness against all kinds of noise interference,signal processing manipulations and malicious attacks.Finally,a semi-fragile zero watermarking algorithm is designed based on the high frequency property of audio signal.This algorithm successfully combines the vantages of digital signature technology with the vantages of traditional semi-fragile watermarking technology in the integrity verification.Experimental results demonstrate that this algorithm achieves good semi-fragility,detection reliability and tampering localization property.(2) First,an echo detection algorithm based on the autocorrelation of power cepstrum is proposed and proved mathematically.Thanks to period invariability and random noise suppression property of autocorrelation function,this algorithm achieves higher echo detection rate and greater robustness than traditional algorithms.Then,in view of the merits of echo hiding technique of simple algorithm,good transparency,undemanding synchronizing,great robustness and blind extraction,a multiple watermarking scheme based on echo hiding is designed and blind extraction of multiple watermarks is also achieved.Experimental results demonstrate this scheme has great robustness against various signal processing manipulations and malicious attacks.(3) To achieve the transparency,avoid the mutual interference between watermarks of different properties and realize the independent and blind extraction of watermarks of different properties in audio multipurpose watermarking system,a new design idea that generates the robust watermark based on "zero watermarking" idea and utilizes the instantaneous mixing model in blind source separation system to embed the fragile watermark in the high-frequency components of of original audio signal is proposed.And,based on this design idea,a multipurpose watermarking scheme based on the vector quantization of low-frequency components of dual-transformation domain and independent component analysis and another scheme based on discrete wavelet transform and principal component analysis are proposed,respectively.Experimental results demonstrate that these two schemes achieve better transparency than other audio multipurpose watermarking schemes.They also achieve good robustness, detection reliability and tampering localization property.(4) To realize the copyright protection and pirate tracking of digital audio works in network,a study is made of the application system of digital audio watermarking.First,a multiple watermarking scheme based on two-stage vector quantization of low-frequency coefficients of dual-transformation domain is put forward.Different methods are utilized to embed two watermarks in the indices of the first stage and the second stage vector quantization of the low-frequency coefficients of dual-transformation domain of original audio signal,respectively.Then,a copyright protection and pirate tracking model based on the proposed multiple watermarking scheme and the buyer-seller watermarking protocol is designed.In the proposed model,the multiple watermarking scheme is utilized to embed the copyright watermark for identifying the copyright owner and the digital fingerprint for identifying the buyer in the digital works,simultaneously.Once a piratical work appears, the watermarks are extracted to identify the copyright violator and to resolve the copyright dispute.At last,the main contributions of this dissertation are summed up and the research areas worthy further studying are summarized.