基于采样点倒置和BP网络的音频掩密算法  

Audio information hiding algorithm using sample dots inversion and BP network

作　　者：杨艳秋[1] 李建勇[2] 曹长修[1] 

机构地区：[1]重庆大学自动化学院,重庆400030 [2]重庆通信学院四系,重庆400035

出　　处：《华中科技大学学报（自然科学版）》2009年第12期13-15,共3页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(60670215);重庆市自然科学基金资助项目(CSTC2007BB2105)

摘　　要：为了更好地提高音频数字水印的鲁棒性和不可感知性的平衡,利用神经网络的自学习和非线性映射能力,提出一种基于采样点倒置和反向传播神经网络的音频水印算法.算法首先对载体音频进行分帧,利用倒置的方法在音频中嵌入秘密信息;再利用BP算法对含密音频和载体音频建立人工神经网络,接受端利用该网络提取秘密信息.实验表明,该算法不仅具有良好的不可感知性,能够抵抗诸如低通滤波、加噪声、回声、重采样、MP3解压缩等攻击,特别还能抵抗DA/AD转换和各种去同步攻击.To improve the better balance between robustness and imperceptibility for audio digital watermark used fully the flexibility and the compatibility of neural network, an audio watermark algorithm based sample dots inversion and back-propagating neural network was proposed. The covert audio was divided into several frames, and then the audio sample dots were inverted to embed secret information; the algorithm establishment artificial neural networks were built by the algorithm using the back propagation （BP） algorithm to the watermarked audio, and the watermark information was extracted used the network to accept. The results show that the method yields a high recovery rate after attacks by commonly used audio data manipulations such as low-pass filtering, re-quantization, re-sampling, echo, MP3 compression, especially it can resist the attacks of the DA/AD conversions and asynchronies.

关 键 词：信息隐藏 反向传播神经网络 数字水印 鲁棒性 去同步攻击 采样点倒置 

分 类 号：TN912.3[电子电信—通信与信息系统]