通信学报 ›› 2015, Vol. 36 ›› Issue (1): 121-128.doi: 10.11959/j.issn.1000-436x.2015014

• 学术论文 • 上一篇    下一篇

基于变分贝叶斯学习的音频水印盲检测方法

唐鑫1,马兆丰1,2,钮心忻1,杨义先1   

  1. 1 北京邮电大学 信息安全中心,北京100876
    2 北京国泰信安科技有限公司,北京 100086
  • 出版日期:2015-01-25 发布日期:2017-06-21
  • 基金资助:
    国家自然科学基金资助项目;国家自然科学基金资助项目;国家自然科学基金资助项目;国家自然科学基金资助项目

Blind audio watermarking mechanism based on variational Bayesian learning

Xin TANG1,Zhao-feng MA1,2,Xin-xin NIU1,Yi-xian YANG1   

  1. 1 Information Security Center,Beijing University of Posts and Telecommunications,Beijing 100876,China
    2 Beijing National Security Science and Technology Co.,Ltd.,Beijing 100086,China
  • Online:2015-01-25 Published:2017-06-21
  • Supported by:
    The National Natural Science Foundation of China;The National Natural Science Foundation of China;The National Natural Science Foundation of China;The National Natural Science Foundation of China

摘要:

为了提高音频水印的检测性能,基于音频帧MFCC特征的统计特性,提出了一种音频水印盲检测方法。在音频帧的DCT系数上嵌入扩频水印,对嵌入水印的音频帧和原始音频帧分别提取MFCC特征进行训练,分别建立高斯混合模型,并通过变分贝叶斯学习方法估计出高斯混合模型的参数,检测时依据最大似然的原则。实验结果显示提出的方法在音频信号受到噪声干扰和恶意攻击的情况下,相对基于EM算法的方法在误检率上有明显降低,在小样本训练情况下具有更好的效果并且可以有效避兔过拟合的问题。

关键词: 高斯混合模型, 音频水印, 盲检测, 过拟合

Abstract:

In order to improve the performance of audio watermarking detection,a blind audio watermarking mechanism using the statistical characteristics based on MFCC features of audio frames was proposed.The spread spectrum watermarking was embedded in the DCT coefficients of audio frames.MFCC features extracted from watermarked audio frames as well as un-watermarked ones were trained to establish their Gaussian mixture models and to estimate the parameters by vatiational Bayesian learning method respectively.The watermarking was detected according to the maximum likelihood principle.The experimental results show that our method can lower the false detection rate compared with the method using EM algorithm when the audio signal was under noise and malicious attacks.Also,the experiments show that the proposed method achieves better performance in handling insufficient training data as well as getting rid of over-fitting problem.

Key words: Gaussian mixture model, audio watermarking, blind detection, over-fitting

No Suggested Reading articles found!