【作者】 黄昊；

【导师】 郭立；

【作者基本信息】 中国科学技术大学 ， 电路与系统， 2008， 博士

【摘要】 数字化与网络化带给了社会和生活深刻的变化和影响,数字媒体的认证与保护日益迫切。因此数字信息隐藏技术引起了广泛的关注和研究,其中抗压缩编码的音频水印具有很高的实用价值。本文的工作主要致力于应用于音频感知编码的信息隐藏算法的设计,力图从理论到实践探索出一种适用于主流感知编码的音频信息隐藏方法。以此为目标本文主要工作及创新点如下:1.通过实验探讨了扩频调制中相关检测准确率降低的根本原因——短时音频信号样点不吻合高斯正态分布模型假设。针对这一问题提出了一种新的自适应的扩频调制方法,应用到音频水印中去。通过调整嵌入强度消除了嵌入域样点与扩频序列相关性的影响,同时通过检测期望的嵌入帧筛选策略来对不可感知性进行优化。2.基于听觉感知特性,提出了新的基于梅尔倒谱系数方差和与连续子帧相关和的感知敏感成分划分法。利用此方法提出一种通过对语音信号进行非均匀压扩来提升时长规整的感知质量的新方法和一种新的针对感知敏感成分的音频信息隐藏的同步攻击方法。3.探讨了隐藏信息与载体信号的相关性对音频感知质量的影响。利用嵌入的非相关提出了一种新的针对直接扩频隐写方式的音频隐写分析方法,根据原始嵌入与检测嵌入引起失真的区别,采用音频失真测度作为特征向量对支持向量机进行训练分类,检测中不需要对原始音频进行估计。4.从感知编码原理出发,提出了抗感知编码水印设计的关键在于要适应于载体感知特性的观点。在音频信号感知规律与时频结构的分析基础上,全面探索了感知编码对隐藏信息的影响。最终提出了一种新的鲁棒音频水印算法。采用载体信号划分的同步机制,利用低频能量的鲁棒性,对相邻子帧的低频能量比进行量化调制来嵌入水印。并对其进行了同步攻击与stirmark攻击测试。更多还原

【Abstract】 The rapid development of digital techniques and internet brings urgent challenges upon multimedia copyright protection and authentification. By the popular demand, watermarking technique attracts extensive attention with plenty of research. This dissertation presents efforts in both theory and practice, on robust digital watermarking for perceptual audio coding, which takes on a significant utility value. The research mainly leads to the following innovations:1. Perturbation in correlation detecting is explored that short-term signals’ actual distribution can’t fit the hypothesis of Gaussian Normal model. So a new self-adaptive spread spectrum audio watermarking is proposed by embedding intensity modification to eliminate the correlation between audio signals and m sequence, with perceptual optimization using embedding frame sift.2. Based on human acoustic features, a new method to divide perceptually critical segments by MFCC and correlation is proposed, which is used in TSM as an improvement and also an effective synchronization attack method to audio watermarking.3. The impact on audio perception of watermarking with little correlation to audio signal is discussed, which leads to a new method of spread-spectrum steganalysis based on distortion measures. The support vector machine (SVM) is used as a classifier ,using various distortion measures sensitive to DSSS stego-method as features, avoiding estimation of the original signal.4. On the foundation of perceptual coding principles, the compression distortions in both time and frequency domain are comprehensively considered. By taking advantage of the robustness of low frequency energy, a watermarking method of quantizing neighbored frame is proposed.更多还原