摘要
为了取得鲁棒性与不可感知性的良好平衡,提出一种新的SVM数字音频水印嵌入算法.该算法以回归型支持向量机(SVR)理论及同步码技术为基础,首先结合数字音频内容,选取稳定的能量、过零率作为特征向量并获得SVR训练模型,然后利用SVR训练模型自适应确定嵌入位置,最后采纳量化调制策略,分别在空间域和频率域内将同步码和数字水印嵌入到音频载体内.仿真实验表明,本文算法不仅具有较好的不可感知性,而且对诸如叠加噪声、重新采样、重新量化、低通滤波、MP3压缩、随机剪切等攻击均具有较好的鲁棒性.
On the basis of support vector regression (SVR) and synchronization code, a new digital audio blind watermarking algorithm is proposed. Firstly, the energy and zero-crossing feature are selected as train sample, which are extracted from the audio segments, and a SVR model can be returned through the SVR train function;then,the watermark embedding position (the significant audio segments) can be obtained adaptively by using the SVR model. Finally, synchronization code and digital watermark are embedded into the significant audio segments in spatial and DWT domain by using quantization modulation. Experimental results show that the proposed audio watermarking scheme is not only inaudible,but also robust against various attacks (such as noise adding,re-sampling,re-quantization,filtering,MP3 compression,and random cropping et al.
出处
《小型微型计算机系统》
CSCD
北大核心
2008年第9期1731-1735,共5页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(60773031)资助
信息安全国家重点实验室(中国科学院软件研究所)开放基金项目(03-06)资助
大连市科技基金项目(2006J23JH020)资助
计算机软件新技术国家重点实验室(南京大学)开放基金项目(A200702)资助
"图像处理与图像通信"江苏省重点实验室(南京邮电大学)开放基金项目(ZK205014)资助
江苏省计算机信息处理技术重点实验室(苏州大学)开放课题基金项目(KJS0602)资助
关键词
音频水印
回归型支持向量机
同步码
量化调制
audio watermarking
support vector regression (SVR)
synchronization code
quantization modulation