Research And Realization Of The Robust Digital Audio Watermarking Scheme

The digital media has become a main way for information communication along with the rapid development of digital technology and computer networks. Protection of digital multimedia information has become an increasingly important issue. Traditional information security system can only safeguard information transmitting process, but can't control the decoded media data. So it can't prevent the illegall copy of the pirate. As a novel way to solve these problems, digital watermarking technology begins to be popularly researched and used. By embedding some secret watermark information in the host multimedia signals, it provides solutions to copyright protection and content verification.Firstly, this paper introduces the development and some related concepts of digital watermarking, which elucidates the background and significance of the research. Secondly, the fundament and general framework of watermark are stated and basic demands are listed. Then, two audio watermarking schemes are presented. A meaning digital audio watermarking scheme based on psychoatic in wavelet domain is presented and simulated. The watermark, a visually recognizable 64x64 binary image that is chaotically encrypted, is embedded into the low frequency coefficients of wavelet transformation of audio signal based on quantization. According to the masking characteristics of the human auditory system, psychoacoustics model is used to compute the quantization step size, so that the watermark is imperceptible by the user and is robust to signal manipulations. Especially, it can be against the cropping and MP3 compression attacks for being embedded synchronization code. A novel audio watermarking scheme with neural network is proposed. The pseudorandom permution watermark sequence and chaotic sequence are combined to the embedding watermark templet sequence, and the original audio signal can be divided into the synchronization module and watermark module. Then the synchronous code is embedded into the synchronization module by quantization in the time domain, and the hiding watermark sequence is embedded into the middle and low frequency coefficients of Discrete Cosine Transformation of watermark module. Extraction processing of the watermark contains three steps, which are determining the synchronous code situation firstly, training the neural network secondly with the sample of the watermark module corresponding to chaotic sequence of the watermark templet sequence until reaching the convergence and lastly extracting the watermark sequence by the trained neural network fromthe validation sample of the watermark module. Finally, the aspects for the future research and the develop trends of digital watermarking are discussed.Experimental results show that the two schemes can preserve inaudibility and they are robust enough to against the different signal processing operations. In addition, very large watermarking information can be embedded into the original audio signal. Furthermore, it can resist the cropping and MP3 compression attacks. So it can be used for copyright protection of digital audio productions.