【作者】 吴明巧；

【导师】 金士尧； 朱中梁；

【作者基本信息】 国防科学技术大学 ， 计算机科学与技术， 2007， 博士

【摘要】 随着网络和信息技术的发展,数字多媒体数据可以在通信网络中迅速便捷的传输,带来通信便利的同时,也给信息安全带来新的挑战。信息隐藏技术成为信息安全领域的研究热点。作为信息隐藏技术主要分支的信息隐写技术,通过将秘密信息隐藏到公开的数字媒体中,以达到传输秘密信息的目的,因而具有更强的隐蔽性。研究信息隐写技术及隐藏信息的检测—隐写分析技术,对国家信息安全具有重要意义。本文主要研究以数字图像为载体的信息隐写及隐写分析技术。论文对图像信息隐写技术进行了回顾,利用信息论对信息隐写系统的安全性进行了分析。提出了一种DCT域的图像信息隐写算法。该算法将数字图像置乱技术中的幻方变换与数字通信中的数字调制技术相结合,以二值文字图像为待隐藏的秘密信息,以静止图像为载体。实验证明算法透明性好,具有良好的抗剪切性。在深入研究人类视觉系统模型的基础上,提出了基于视觉模型的小波域数字图像隐写算法。考虑人类视觉系统的特性,给出图像小波域临界噪声域值矩阵JND(justnoticed difference),利用迭代混合的方法将秘密信息白适应的嵌入原始图像的小波系数中。实验证明该算法具有良好的透明性和鲁棒性。论文对图像隐写分析技术进行了回顾,总结了图像统计模型。图像隐写工具Jsteg使用图像DCT系数的LSB隐藏秘密信息,传统的卡方攻击只能检测序列Jsteg隐藏,不能检测随机Jsteg隐藏。提出了一种基于图像统计模型的针对Jsteg隐写分析算法。算法利用Laplacian分布拟和图像AC系数的统计量,运用皮尔逊x~2进行拟和度检验,使用线性回归估计嵌入量。该方法能够检测序列Jsteg嵌入和随机Jsteg嵌入。隐写算法中利用图像DCT系数最不重要位隐藏信息的隐写算法具有透明性好,鲁棒性强等特点。这类隐写算法包括Jsteg,Outguess等。论文提出了针对DCT系数最不重要位隐写算法的隐写分析算法。算法利用DCT系数的相关性,以统计测试值作为特征,运用支撑向量回归判别隐秘图像与原始图像。实验证明该隐写分析算法能够有效检测Jsteg隐写和Outguess隐写。主动隐写分析的目的不光是判别隐秘图像,而且估计嵌入算法的参数为秘密信息的提取作准备。论文提出了一种针对图像扩展频谱序列隐写的密钥估计算法。该算法基于序列分析与突变检测的理论,考虑图像DCT系数满足拉普拉斯分布,给出了理想平稳拉普拉斯分布信号扩展频谱隐藏密钥估计的模型。对于非平稳信号的图像数据,利用局部最有效序列检测法,给出了拉普拉斯分布的密钥估计模型。实验证明,该方法不但能检测出扩展频谱隐写,估计嵌入密钥,而且比Trivedi的方法更有效。设计实现了数字图像隐写分析平台。该软件平台基于VC++语言实现,可对BMP图像和JPEG图像挂接隐写、空域隐写、频域隐写及部分隐写软件工具隐写进行检测。介绍了文本信息隐写技术。深入分析了文本隐写工具Stego的隐写原理,提出了针对Stego的隐写分析方法。当隐写所使用的字典单词全以小写字母开头时,可以通过基于标志特征的隐写分析方法检测出隐秘文本;当上述条件不满足时可通过基于统计特征的隐写分析方法检测出隐秘文本。更多还原

【Abstract】 With the development of network and information technology, digital media data can be transmitted fleetly and conveniently on communication netwok. Bring conveniency to communication, it also brings new challenge to communication security. Information hiding has become the hotspot in reseach of communication security. Steganography, as the main embranchment of information hiding, hiding the secret information into open digital media for the purpose of secret communication, is more imperceptible. Research on steganography and the detection of hiding message is very important to national communication security. This thesis is mainly researching on steganography and steganalysis technique of images.A review of steganorgraphy in images is given. The security of steganography system is analysised using information theory. A steganography algorithm in DCT domain of image is proposed. Using literal image as secret message and still image as cover, this algorithm combines image scrambling with digital modulation. Experiment results show that the property of invisibility and anticropping is good. Based on human visual system model, a novel image steganography algorithm in wavelet domain is proposed using the iterative blending method. Giving the character of human visual system, a JND (just noticed difference) threshold matrix is caculated pixel by pixel in the image subbands. Using the iterative blending method, the secret message is embedded into wavelet coefficients adaptively. Experiment results show that the property of invisibility and robustness is good.A review of steganalysis in images is given. Steganography software Jsteg uses the LSB of DCT coefficients to hide secret message. The tranditional Chi-square test can only detec the sequential Jsteg hiding and can not detect the random Jsteg hiding. A new fast steganalysis algorithm for detecting Jsteg hiding based on statistical model of image is proposed. The algorithm uses the Laplacian to fit the distribution of statistic of AC coefficients and uses Pearsonχ~2 test to test goodness-of fit. The amount of embedding message is estimated using linear regression. The algorithm can detect both sequential Jsteg hiding and random Jsteg hiding.The kind of steganography algorithms using the least significant bit of the DCT coefficients to hide message have the characters of invisibility and robustness. This kind of steganography algorithms includes Jsteg, Outguess and so on. A steganalysis algorithm that can detect the hiding in the least significant bit of the DCT coefficients is proposed. The algorithm is based on the thought that the DCT coefficients are correlative. Using the statistical tests as the features, support vector regression is trained to discriminate the stego-images from the clear ones. Experiment results show that our method can detect the hiding by Jsteg and OutGuess. The purpose of active steganalysis is not only detecting stego-image, but also estimate the parameter of embedding algorithm for extraction of the secret message. A steganalysis algorithm that estimates the secret key used in sequential steganography of spread spectrum embedding is proposed. Considering the Laplacian distribution of image DCT coefficients, a secret key estimation model of stationary laplacian host signal embedded by spread spectrum steganography is presented based on the theory of detection of abrupt changes and sequential detection. For non-statioanry digital image, a locally most powerful steganaysis detector is derived based on Laplacian distribution. The results of experiments show that our method is more efficient than Trivedi’s method.An image steganalysis software is designed and implemented by Visual C++ language. This software can detects the steganography of connecting, hiding in spacial domain, hiding in frequent domain and hiding using some steganography software.A sumarise of steganography in test is given. By deeply study the mechanism of Stego, two steganalysis methods are proposed. When the first letters of words in dictionary are lowercases, the steganalysis method based on the signature characters can detect Stego. When it’s not the case, the steganalysis method based on statistical characters can detect Stego.更多还原