乱序局部敏感哈希音频零水印方案

Zero-Watermark Scheme for Audio Using Disorder-Locality-Sensitive Hashing

针对现有音频零水印算法缺乏安全分析性研究、构造过程复杂化、唯一性差、未考虑音频帧信息含量分布不均等问题,提出了一种命名为分析重构攻击的特殊攻击模型,并基于该攻击模型结合音频信号小波特征与局部敏感哈希,设计了一种抗分析重构攻击的音频零水印方案。该方案分为零水印构造与零水印检测这2大模块:在零水印构造模块,首先提出并设计了去静音帧模型,利用该模型对音频信号进行预处理,从根源上降低水印构造过程中的复杂度,再结合链表、局部敏感哈希以及小波变换构造零水印,最后将水印信息与水印图像结合;在零水印检测模块,通过对水印信息的相似度计算与水印图像的相似度比对共同判断版权归属。为了验证方案有效性及可行性,从方案自身验证与对比分析这2方面对方案进行验证分析,理论分析与实验结果表明:该方案所构造的音频零水印最小分析重构复杂度为特征矩阵的平方阶,具有较好的不可感知性、抗噪声攻击性、抗MP3压缩攻击性及抗多比例剪裁攻击性等优势。

Aiming at the problems such as lack of security analysis, complicated construction process, weak uniqueness, and neglect of considering the audio information content distribution, a special attack model named analysis and reconfiguration attack model is proposed. Based on this attack model, a new scheme of audio zero-watermarking is designed, which combines discrete wavelet transform with locality-sensitive hashing. The scheme contains two main modules of zero-watermarking construction and zero-watermarking detection. In the zero-watermarking construction module, a de-mute frame model for signal is designed. This model uses the attack model to preprocess audio signal for reducing complexity of zero-watermarking construction, and constructs watermark by link list, locality sensitive hash and discrete wavelet transform, then the watermark information and watermark images are combined. In the zero-watermark detection module, the copyright attribution is determined by calculating the similarity of the watermark and the similarity of the watermark images. Finally, the effectiveness and feasibility of this new scheme are verified via self-verification and comparative analysis. Theoretical analysis and experimental results demonstrate that this scheme can construct an audio zero-watermark with minimum analysis and reconfiguration complexity of square order of eigenmatrix, and has the advantages of good imperceptibility, robustness, and capabilities of resisting MP3-compression attack and multi-proportional tailoring attack.