基于正交随机空间投影的数字音频扩谱水印

The Digital Audio Spread-Spectrum Watermarking Based on Random Orthogonal Projection

数字音频水印是在原始音频样本中嵌入加密的标志信息以实现音频内容的注释和版权保护。对于任何一种音频水印的算法,必须满足水印的透明性和鲁棒性两个基本要求。提出了一种新的基于正交随机空间投影的数字音频扩谱水印算法,它将音频样本的中频子带重新排列成MxN的矩阵形式,并投影到随机的正交子空间上,然后将扩谱后的水印序列嵌入到相应的投影系数中,为了避免产生听觉上失真,算法中利用心理声学模型控制水印产生的听觉噪声。水印的检测过程可以通过扩谱序列的相关进行盲检测,实验结果表明采用该算法能够很好地保证水印在听觉上的不可觉察性,与现有的数字音频水印算法相比,该算法对数字音频信号的压缩、滤波、加噪声和加入回波等处理具有更好的鲁棒性。

Digital audio watermarking embeds secret information into the original audio to achieoe annotation and copyright protection. With respect to any audio watermarking scheme, it is very important to meet the requirements of transparency and robustness. A novel audio spread-spectrum watermarking scheme based on the random projection is proposed in this paper. By sing the discrete wavelet transform, the middle-frequency sub-band of the original audio signal is selected, and the coefficients are rearranged into the M × N matrix. Subsequently, an orthogonal random projection is performed on this matrix. Spread-spectrum watermarking sequences are then embedded into the coefficients in the projection domain. In order to avoid the auditory distortions, we employ the psychoacoustic model to control the auditory noise generated by the watermark. The detection process is the blind auto-correlation operation without referring to the original audio. Experimental results have shown that the embedded watermark is inaudible, and compared with the existing watermarking schemes, it is more robust against the audio compression, filtering, noise adding and echoing operations.