Study On Audio Watermarking And Analysis On Distortion Law For Embedding Watermark

Sound is an important way to transfer information for mankind. With the development of communication networks and electronic products, the technology of processing audio information has been widely used. How to identify the copyright of audio works, track pirated audio works, and judge the originality and integrity of audio data? One of the ways to resolve these problems is digital watermarking.The author researches the audio watermarking technology which is used to protect copyright of audio works and authenticate the integrity of audio works. At the same time, the distortion laws of audio samples, due to modifying coefficients in some transform domain to embed watermark, are analyzed. The achieved results are listed below.1. Based on neighborhood averaging method, a robust and high efficient synchronization scheme is proposed. In addition, a robust audio watermarking algorithm is presented, and then the inaudibility of the watermarked audio is improved. Experimental results show that the synchronization scheme has strong robustness, and the imperceptibility of the watermarked audio is excellent. In addition, the watermarking algorithm is robust to not only common signal processing such as noise adding, re-sampling, re-quantization, low-pass filtering, but also de-synchronization attack such as slight time-scale modification, random cropping and jittering attacks.2. Neighborhood averaging sequences is the low-frequency component of an audio signal, and DCT(discrete cosine transformation) has the energy-compression characteristics. Combining neighborhood averaging method and DCT, a robust audio watermarking scheme is proposed. Subjective and objective tests reveal that the proposed watermarking scheme maintains highly audio quality and the synchronization mark can be searched quickly and accurately, and the algorithm is simultaneously highly robust to common digital signal processing attacks, including additive noise, sampling rate change, bit resolution transformation, MP3compression, and random cropping, especially to low-pass filtering operation.3. Even quantitation step is very small, the watermark can be extracted when it was embedded in neighbor averaging sequence. The time-domain waveform envelope can be almost preserved due to scaling amplitudes in a group manner by the same little amount. Based on this property, a fragile audio watermarking scheme for content authentication was proposed. Subjective and objective tests show that the fragile watermarking scheme is inaudible and sensitive against MP3compression, the noise adding, re-sampling, re-quantization, low-pass filtering, etc. Especially, it can detect and localize the tampered regions effectively.4. To a one-dimensional sequence, the modification of alternating coefficient in DCT domain will not change amplitude sum of the sequence in time domain. Based on this property, a content-based fragile watermarking algorithm was proposed to realize the integrity authentication of audio files. Experimental results show that the fragile watermarking algorithm maintains good perceptual quality, the SNR (signal-to-noise ratio) values of watermarked audios are about80dB, and the algorithm is sensitive against additive noise, bit resolution transformation, sampling rate change, low-pass filtering, MP3compression. In the meantime, it can detect and localize the tampered regions effectively.5. The distortion laws of audio samples are obtained when SVD(Singular Value Decomposition), DCT, DWT(Discrete Wavelet Transform) or DWT-DCT hybrid domain, which are widely adopted to embed the watermark information into audio signal. For the SVD, after a set of audio samples is rearranged to2-dimensional matrix, the SVD is performed. Then the Euclidean norm of all singular vectors can be computed from all singular vectors. The audio samples are expanded or reduced by the same proportion when the Euclidean norm is modified. For DCT, the distortion of audio samples, due to modifying a certain alternative coefficient in DCT domain, is the cosine curve after inverse DCT. For DWT, distortion of audio signal appears the shape of scaling function after inverse DWT if a certain approximation coefficient in DWT domain was justified. Furthermore, analysis on embedding watermark in DCT-DWT hybrid domain is performed. At the same time, experiments are done to verify these laws. Our conclusions are helpful to find the discontinuities due to embedding watermark, which may cause audible noise, and improve the efficiency of embedding watermark.