基于直方图特性的时域鲁棒音频水印算法

鲁棒音频水印算法应该能够抵抗多种典型攻击，但目前具有这种综合性能的算法还很少。在详细分析时域数据受到攻击前后的分布基础上，发现了数据统计均值呈现出良好的不变特性，且数据直方图4个连续Bin样本关系总是保持在±50之内。提出了基于概率统计特征的通用数据选择方法，适用于正态分布数据的水印隐藏要求。然后，利用直方图4个连续Bin样本的稳定关系原理，设计了基于统计特征的水印嵌入和提取算法。实验结果表明：该算法除了能抵抗MP3、重采样、重量化和规一化攻击外，还完全能够抵抗4k低通滤波、1050至70％的基频不变TSM、以及10％和200％的重采样TSM攻击，而且在10％至250％的范围内容完全抵抗音量缩放攻击。此外，建立的步长调配因数有助于进一步提高隐藏效果。

Robust Audio Watermarking Algorithm Based on Histogram Feature

Few audio watermarking algorithms are robust to most of the audio attacks today. Based on the data distribution in time domain before and after attacking on the audio, it is clear that the mean shows good invariant statistical feature. The relation of four consecutive bins in a histogram keeps within ±5%. Then, a general data selection approach, which meets the demands of normal distribution data, is created for the watermark hiding. Furthermore, an audio watermarking algorithm of embedding and extracting is designed with statistical features of the stable relation of four consecutive bins. Experimental results show that the algorithm can resist on attacks of low pass filtering to 4 kHz, pitch invariant TSM changes from 105% to 70%, and resample TSM changes from 10% to 200%. The algorithm can also resist on the attack of amplifying change of ±30%, along with strong robustness to the common attacks of MP3 compression, re-sampling, re-quantizing and normalizing. Besides, a modification factor is established for the embedding step to get higher hiding performance.