【作者】 谢建全；

【导师】 阳春华；

【作者基本信息】 中南大学 ， 计算机应用技术， 2009， 博士

【摘要】 互联网络和信息科学的飞速发展,为信息的传输和处理带来了极大方便,但是网络在给人们带来便利的同时也暴露出越来越严重的安全问题。信息隐藏作为一种保证信息安全传递的重要手段,引起了人们的极大关注,它通过隐藏信息的“存在性”来保证信息的安全传输,能解决密码学在应用上容易引起攻击的问题。信息隐藏的应用包括隐秘通信、版权保护、数字指纹、多媒体内容篡改认证等军事和民用领域,有重要的理论和应用研究价值。在过去的十多年中,信息隐藏的一大分支数字水印取得了快速的发展,但在隐秘通信和篡改认证方面取得的研究成果相对较少,其主要原因是缺乏对安全性和感知质量客观准确的评价手段,隐藏算法的嵌入容量、安全性、感知失真之间的固有约束存在着一定的难度,尤其是多数隐藏算法的嵌入容量还远未达到隐秘通信的容量要求。本论文的主要工作就是围绕如何提高信息隐藏容量,以满足隐秘通信和篡改认证等应用所需的嵌入容量这个目标展开的,内容涉及信息隐藏的不可感知性评估、数字图像的加密与置乱、大容量的图像信息隐藏算法等方面。论文的主要研究成果与创新性如下:(1)分析了峰值信噪比(Peak Signal to Noise Ratio,PSNR)、像素块误差分布(Pixel-Block Error Distribution,PBED)等一些目前常用的视觉不可感知性评价指标的不足之处,根据人类视觉特性提出了一种衡量信息隐藏算法不可感知性的视觉失真感知函数(VisionDistortion Sensitivity Function,VDSF),它反映了人类视觉的四个特性:①人眼对视觉信号变化剧烈的地方(纹理区)的噪声不敏感,对平滑区域的噪声敏感;②人眼对水平方向的变化比对角线方向更为敏感;③低于可感知极限的亮度变化是不可感知的;④相邻位置的像素是相关的,并且相关性与距离有关。VSDF指标计算复杂度低,并且PSNR是它的一种特殊情况。实验证明VDSF与人类视觉感知质量主观评价的一致性明显优于PSNR等传统指标,能更好地衡量隐藏信息的不可感知性。实验还证明VDSF也适用于对二值图像的质量评价,解决了灰度图像与二值图像不能使用同一评价指标的问题。(2)研究了基于混沌序列加密和混沌置乱的信息预处理方法,提出了基于Logistic映射的混沌序列产生算法,所产生的序列不再出现Logistic映射所存在的空白窗口和稳定窗等安全问题,并且密钥空间大幅度增加;改进了基于Arnold变换的混沌置乱算法,改进后的算法不再出现Arnold变换的周期性现象,且能有效抵御选择明文的攻击,提高了安全性。(3)分析了图像在空间域的位平面分布特性,指出了直接用秘密信息替换最低有效位(LSB)位平面或多个低位位平面可能破坏相应位平面纹理特性而导致的安全问题,提出了基于最高有效位的位平面分解方法,并得出了最高有效位下的某个低位位平面替换成随机二值图像,不会破坏原始图像在固定位平面分解时低层位平面所固有的纹理特性的结论。利用这一结论和人类视觉的亮度掩蔽效应,提出了两种基于图像空间域的自适应信息隐藏算法,算法有非常高的嵌入容量,同以往的空间域算法相比,能有效防止出现破坏低位位平面纹理特性而导致的安全问题,安全性有很大提高。(4)针对DCT变换域信息隐藏算法虽然有较好的鲁棒性但嵌入容量小的问题,研究了自然图像的频谱分布特性和JPEG压缩不变性,分别提出基于频谱均匀化和在高频系数中嵌入信息的大容量DCT域隐藏算法。基于频谱均匀化的算法通过对载体图像进行置乱处理实现频谱均匀化,以获得更多的可用于隐藏信息的DCT系数来提高算法的嵌入容量。在高频系数中嵌入信息的隐藏算法主要利用JPEG压缩不变性来实现,在保持不可感知的前提下,该算法的嵌入容量大于在中低频系数嵌入信息的算法,而且对不超过预设质量因子的有损压缩有强鲁棒性。(5)根据二值图像非白即黑的特点,提出了基于分块的大容量二值图像信息隐藏算法。通过将二值图像分割成大小为2×2的图像子块,再根据每个子块中黑白像素的个数的不同,自适应地确定可嵌入隐藏信息的比特数及嵌入的位置,每个处于黑白像素交界的子块至少能隐藏2比特信息,嵌入容量比目前同类的盲提取算法高一倍以上。更多还原

【Abstract】 Along with the rapid development of Internet and information science,which are providing great convenience for information transmission and processing,increasingly severe security problems have emerged gradually.Information hiding,as an important method to guarantee secure information transmission,has attracted great attention all over the world.By hiding the existence of information,hiding communication has solved the easily-being-attacked cryptographic problems in application.Its applications include copyright protection, digital fingerprinting,tamper authentication of multimedia,hiding communication,etc.of military or civil use;therefore,it is of great theoretical and practical importance.In the past decade,digital watermarking,as a branch of information hiding,has been developed rapidly.But achievements in hiding communication and tamper authentication are relatively insufficient because there exists some difficulties like inherent constraints among embedding capacity,security, perceptual distortion with the lack of objective measurement to both security and perceptual quality,especially embedding capacity of most present hiding algorithms can hardly reach the demands of hiding communication.The focus of this dissertation is on the study of evaluating the imperceptibility and security of hiding information and designing embedding algorithms with higher hiding capacity,mainly including imperceptibility evaluation of hiding information,encryption and scrambling of digital images,high capacity image information hiding algorithm,and so on.Some of the conclusions are listed as follows:(1) Based on the analysis of the drawbacks of commonly used visual imperceptibility criteria,such as PSNR(Peak Signal to Noise Ratio), PBED(Pixel-Block Error Distribution),a new criterion called VDSF (Vision Distortion Sensitivity Function) is proposed to evaluate the imperceptibility of hiding information algorithm according to human vision properties,which characterizes four aspects of human vision.First, human eyes are not sensitive to noise within violently varying visual signals(such as texture) but sensitive to noise in smoothing area.Second, human eyes are more sensitive to changes in horizontal direction than ones in diagonal direction.Third,brightness variances lower than perceptible limit are imperceptible to human.Fourth,adjacent pixels are correlated with each other,and the relativity depends on their distance. VDSF has lower computational complexity.Especially PSNR belongs to one of its specific cases.Experiments show that VDSF can overcome drawbacks of present visual imperceptibility criteria like PSNR and evaluate better the imperceptibility of hiding information.Moreover,the proposed algorithm is suitable for quality evaluation of binary images to solve the problem that the individual criterion can not be used for both grey image and binary image in the meantime.(2) On the basis of chaotic map information preprocessing,a logistic map based algorithm generating chaotic sequences is proposed,which avoids secure problems such as blank windows,stable window,etc. Moreover,the key space is greatly increased.A chaotic scrambling algorithm based on Arnold map is improved.In this scheme the periodicity of Arnold map is eliminated,it can effectively defend chosen plaintext attack.(3) Bit plane distribution property of an image in spatial domain is analyzed,and secure problems caused by damaging its bit plane texture characteristics due to replacing least significant bit(LSB) or multiple low bit plane with direct secret messages,is discussed.A bit plane decomposition method based on MSB(most significant bit) is proposed. Moreover,a conclusion that replacing certain low bit plane below MSB with random binary image doesn’t damage its inherent texture property when original image is decomposed in fixed bit plane,is drawn.Based on image spatial domain,two adaptive information hiding algorithms are proposed by applying the above conclusion and the brightness masking effect of human visual system.The proposed schemes not only have high capacity but also are effective to protect against damaging texture property of low bit plane so as to improve the security.(4) DCT(Discrete Cosine Transform) domain based information hiding algorithm has merits of good robustness but weakness of low embeddable capacity.Based on spectrum analysis of natural images,high capacity DCT hiding information algorithms of both frequency spectrum homogenization and embedding information in high frequency coefficients are presented respectively.In frequency spectrum homogenization algorithm embeddable capacity is improved by using frequency spectrum homogenization processing on original image to obtain more DCT coefficients used for hiding information.Another algorithm uses JPEG invariance property,and information is embedded in high frequency coefficients under the conditions of imperceptibility in order to improve embeddable capacity.Moreover,this method has good robustness to lossy compression within default quality factors.(5) High capacity hiding information algorithms in binary host image is developed according to characteristics of binary image.The binary host image is first divided into blocks of size 2×2,and then each sub-block adaptively determines the embedded bits number and their positions according to the difference between its black pixel and white pixel number.Each sub-block in the transitional block between black pixel and white pixel can hide at least two bits messages.Therefore,its embeddable capacity is two times more than that of other presented blind algorithms.更多还原