摘要
提出了一种基于对象传播神经网络的音频水印算法。算法将水印的嵌入和提取转换为对象传播神经网络(CPN)的训练和回想,由于水印的提取依赖于CPN输入的统计特性,因此选用具有较强稳定性的小波低频系数方差作为输入向量训练CPN。实验结果表明,该算法在抵抗常规音频信号处理和去同步攻击方面具有较好的鲁棒性。
A novel CPN-based audio watermarking algorithm was proposed. The watermark embedding procedure and extracting procedure were integrated into the proposed CPN. Since the watermark extraction depends on the stable statistical properties of the input of CPN, the variance of low frequency wavelet coefficients with high stability servers as the input vector of CPN in the algorithm to preserve the watermark from various common signal processing and desynchronization attack. Experiments validate the robust of the proposed scheme.
出处
《重庆邮电大学学报(自然科学版)》
北大核心
2009年第1期95-99,共5页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
关键词
数字水印
音频水印
对象传播神经网络
去同步攻击
digital watermarking
digital audio watermarking
counter-propagation neural network
desynchronization attack