压缩域语音感知哈希认证算法及安全性研究

发布时间：2018-02-27 02:12

本文关键词： 语音内容认证感知哈希压缩域 MDCT频谱熵安全性分析传输信道安全性数字水印　出处：《兰州理工大学》2014年硕士论文　论文类型：学位论文

【摘要】：随着网络通信的开放和多媒体处理技术的飞速发展,音频信号在传输过程容易遭受到非法的攻击。语音作为一种最为常见的音频形式,在日常通信、法庭取证、指令密保等方面应用广泛,其内容的安全性和完整性验证至关重要。感知哈希的本质是一类数字摘要技术,因其对内容保持操作具有较好的鲁棒性,对非法篡改操作具有一定的区分性,同时也满足单向性、抗碰撞性等基本安全性要求,逐渐成为一种用于图像、音频和视频等多媒体安全认证的热门技术。语音信号与宽带音频在信号特点、编码、传输方式以及存储方面有很大的不同。而传统的音频感知哈希认证算法主要针对原始域宽带音乐,对压缩格式音频和语音适用性较弱。针对以上问题,提出了一种基于MDCT频谱熵的压缩域语音感知哈希认证算法。算法首先通过对MP3格式的语音文件进行部分解压以提取MDCT系数,以子带划分的方法计算MDCT频谱能量来降低运算量。其次将MDCT频谱子带能量映射为与信息熵理论中相似的质量函数,以此作为感知特征集进而量化为二进制感知哈希值。实验结果证明：算法对内容保持操作的鲁棒性有所提高,同时保证了算法的执行效率。该算法由于仅对MP3文件部分解码,算法的实时性很高,能够满足实时通信终端的要求。针对音频感知哈希安全性分析和评价方法被忽视的研究现状,提出一种基于频谱熵特征的压缩域音频水印嵌入算法。在构造感知哈希值的过程中引入密钥并对生成的感知哈希值进行纠错码预处理,将编码之后的数据作为认证信息嵌入载体,达到解决感知哈希值传输过程中需要额外信道以及容易受到攻击而发生改变的问题。实验结果表明：该算法在不影响感知内容的前提下进行嵌入,对于常见攻击具有鲁棒性。最后结合图像感知哈希安全性研究方法,以本文提出的算法作为实验对象,根据实验结果分析当前音频感知哈希算法的安全性,并得出实验结果：当前安全性主要依赖于密钥和针对某项局部内容攻击进行专门的优化。
[Abstract]:With the opening of network communication and the rapid development of multimedia processing technology, audio signal is vulnerable to illegal attack in the course of transmission. Instruction secrecy is widely used in many fields, so the security and integrity verification of its content is very important. The essence of perceptual hashing is a kind of digital abstract technology, because it has good robustness to content-keeping operations, distinguishes illegal tampering operations, and meets the basic security requirements such as unidirectional, anti-collision, etc. It has gradually become a popular technology for multimedia security authentication, such as image, audio and video. Speech signal is different from broadband audio in signal characteristics, coding, transmission and storage, while the traditional audio perceptual hash authentication algorithm is mainly aimed at the original broadband music. This paper proposes a speech perceptual hash authentication algorithm based on MDCT spectrum entropy for audio and speech in compressed format. Firstly, the MDCT coefficients are extracted by partial decompression of speech files in MP3 format. The energy of MDCT spectrum is calculated by subband partition to reduce the computation cost. Secondly, the energy of subband of MDCT spectrum is mapped to a mass function similar to that in information entropy theory. The experimental results show that the algorithm is more robust to the content-preserving operation, while ensuring the efficiency of the algorithm. The algorithm is only partially decoded to the MP3 file, and the algorithm is used as the perceptual feature set and quantized to binary perceptual hash value. The algorithm is very real-time and can meet the requirement of real-time communication terminal. The current situation of the research on the analysis and evaluation of audio perceptual hashing security is neglected. This paper presents an audio watermarking embedding algorithm in compressed domain based on spectrum entropy feature. In the process of constructing perceptual hash, the key is introduced and the generated perceptual hash is preprocessed by correcting code. The encoded data is embedded as the authentication information carrier, In order to solve the problem that the additional channel is needed and vulnerable to attack in the process of perceptual hash transmission, the experimental results show that the algorithm is embedded without affecting the perceptual content. Finally, the algorithm proposed in this paper is taken as the experimental object, and the security of the current audio aware hash algorithm is analyzed according to the experimental results. The experimental results are as follows: the current security mainly depends on the key and the special optimization for a local content attack.
【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN912.3

【参考文献】