基于(t,n)门限的自同步小波域音频盲水印技术

Self-synchronization blind audio watermarking based on (t,n) threshold

提出一种基于(t,n)门限的自同步小波域音频数字盲水印技术,水印嵌入过程是在小波变换域中进行的.按照秘密共享技术将水印信号分解为n份水印影子,只有t个或大于t个用户才能同时恢复水印,而t-1或更少的用户均不能恢复水印,在增加信息安全性的同时可防止信息泄漏.利用语音信号小波变换自身特点,根据小波变换系数进行基音周期检测,并以此作为水印过程的同步点,水印检测前先搜索同步点,实现水印检测和嵌入过程的自同步.检测过程采用独立分量分析技术,在不需要任何原始音频、水印、嵌入过程信息以及可能经历的攻击信息的情况下可精确恢复水印,实现盲检测.为防止欺骗攻击,在信息恢复前,利用单向散列函数抵抗欺骗攻击.该算法中水印采用的是具有实际意义的语音信号.实验表明,该算法具有很好的鲁棒性.

A self-synchronization blind audio watermarking based on （t, n） threshold in wavelet transform is proposed in this paper. This technique can extract the watermark without using the original watermark and the host audio during the detection. It is a completely blind method during watermark detection. Watermark was divided into n shadows signal according to secret sharing scheme before the embedding process, and Shamir＇s （t, n） threshold scheme was used to distribute the watermark shadows to the users. Lagrange interpolation polynomial was used to realize it. Results show that t or more of those signal shadows can reconstruct the secret watermark, while t - 1 or less signal shadows could not do it. The watermark used in this paper is meaningful speech and it was produced independent of the original audio. The watermark embedded intensity depended on the masking effect of human audio system. Audio processing such as time shift and time scaling modification can cause displacement between embedding and detection process and it is hence difficult for watermark to survive. The pitch detection is done using the coefficients of wavelet transform of the host audio, which is used as the synchronization point. Before watermark detection, the detector finds the synchronization point to realize the synchronization between the watermark embedding and detection process. Recently, blind source separation by independent component analysis （ICA） has received attention because of its potential applications in signal processing. ICA is adopted during the watermark detection to realize the true blind detection without any information about the host audio, watermark, embedding information and attacks. The accuracy of the watermarking extraction also depends on the statistical independence of original audio and watermark signal. To resist the cheating attacks, we use the one way hashing function before watermark reconstruction. Only t or more honest users can work collectively to reconstruct the secret, while dishonest users could not do it. The detailed performance analysis of the proposed watermarking method is presented. Different audio signals are used to test the robustness of the method. Experimental results show that the proposed audio watermarking technique is robust against many audio processing performed by popular watermark test software-Stirmark, such as low-pass filtering, median filtering, additive Gaussian noise, mp3 compression, time scaling, time shift, dithering and quantization.