摘要
隐写分析是信息隐藏检测的重要分支,分为针对性隐写分析和广义隐写分析。针对性隐写分析检测率高,但考虑到隐写术的多样性,它的实际应用受到限制。广义隐写分析适用性强,可以针对任意隐写方法训练,因此对其开展研究具有重要意义。
图像广义隐写分析类似于两类模式识别问题,其目的是为了将给定图片分为隐写和未隐写两类,因此特征提取、特征空间优化和分类器设计是其中的三大关键技术,本文从这三方面进行了深入研究和大量实验,主要内容包括:
1.针对目前广义隐写分析多从灰度图中提取特征而忽略了颜色信息的缺陷,在分析和比较小波子带系数高阶概率密度函数(PDF)矩和直方图特征函数(CF)矩的基础上,提出结合颜色相关度和直方图CF矩的方法。对图像亮度分量引入小波子带直方图CF矩,对色度分量提取矢量方向相关度特征。该方法不但克服了单纯从灰度图中提取特征的片面性而且通过引入直方图CF矩获得了更加有效的隐写分析特征集。实验结果表明,该算法对多种隐写方法均有较好的检测效果。
2.考虑到提取的特征中有些并不能很好地反映由于隐写所带来的统计特性变化,提出利用方差分析对特征进行评估的策略,进一步从中选取更有效的特征以改善分类性能。实验结果表明,经过方差分析特征选择后,隐写分析的特征空间得到有效优化,分类器训练时间大大减少,同时检测率也得到有效提高。
3.由于当前广义隐写分析多为两类隐写分析,只能判断隐写的存在性,因此通过结合直接非循环图支持向量机算法将隐写分析特征集推广到多类广义隐写分析中。这样不仅可以检测图像是否存在隐写还可以推断隐写图像可能使用的隐写工具,为进一步使用针对性隐写分析方法提供了有效信息。比较实验结果表明,该算法具有较好的分类效果,在结合广义隐写分析和针对性隐写分析方面进行了有益尝试。
Steganalysis is an important branch of information hiding detection which can be divided into two categories:specific steganalysis and universal steganalysis.Specific steganalysis has higher detection accuracy but it is not practical owing to the diversity of steganography,while universal steganalysis can be trained on any Steganographic method, so it has a better adaptability.Consequently the research on universal steganalysis is a matter of great significance.
Universal steganalysis is actually similar to pattern classification which centers two-class classification.The goal of it is to divide the image into two groups:cover image and stego-image.So the extraction and optimizing of features and the design of the classifier are three crucial technologies in it.This thesis carries a deep study on those three parts through a great deal of experiments and acquires a series of valuable results which can be summarized in to the following aspects:
First,it is found that recent universal steganalysis schemes usually extract the features in gray images,in this way they neglect the chrominance information of images.On the basis of the comparison and analyzing between high order probability density function moments of wavelet coefficients and characteristic function moments of histogram,a universal steganalysis combining color correlation degree and characteristic function moments of histogram is proposed which extracts the features from both luminance and chrominance components of images.Moments of wavelet characteristic function is introduced to replace the wavelet high-order statistics for luminance component of image and vector directional correlation degree is calculated for chrominance component.This method overcomes the unilateralism of extracting features in grayscale images and improves the effectiveness of feature vector by introducing CF moments.The experimental results have shown that the proposed method offers a better performance over several Steganographic methods.
Second,considering that some features are not able to reflect well the statistical change of the setgo-images.So a feature evaluation strategy based on ANOV(Analysis of Variance)is proposed.By exploring it,the features which are more sensitive to hidden message are selected.The results have shown that after ANOV the dimension of the feature vector reduces to thirty two and the detection accuracy has been effectively improved too.
Third,take into account that most of the recent universal steganalysis schemes are two class steganalysis,so they can only judge the existence of the hidden information.Here we introduce our method into a multi-class steganalysis by using directed acyclic graph SVM(support vector machine) algorithm,This method is not only able to tell whether a image is stego or not,but also can reliably classifies it to its embedding tool.Provide effective information for specific steganalysis.The results have indicated that this method has a better classification result and makes a helpful attempt for combination of universal steganalysis and specific steganalysis.
引文
[1]王朔中,张新鹏,张开文.数字密写与密写分析[M].北京:清华大学出版社,2005.
[2]Niels Provos,Peter Honeyman.Detecting steganographic content on the internet[R].USA:University of Michigan,2001.
[3]Niels F.Johnson,Sushil Jajodia.Steganalysis of images created using current steganography Software[C].Workshop on information hiding,Portland,Oregon,USA,1998,273-289.
[4]Shi Yun Quan,Chen Chun Hua,Gurong Xuan,Wei Su.Steganalysis versus splicing detection[C].Proceedings international workshop on digital watermarking,2008,Guangzhou,China 158-172.
[5]Lou Der-Chyuan,Lin Chih-Lin,Liu Chiang-Lung.Universal steganalysis scheme using support vector machines[J].Optical Engineering,2007,46(11):123-130.
[6]Xiang Yang Luo,Fen Lin Liu,Jiangming Chen,Yining Zhang.Image universal steganalysis based on wavelet packet transform[C].IEEE 10th workshop on multimedia signal processing,Cairns,Australia,2008,780-784.
[7]翟卫东,吕述望,刘振华.基于广义高斯分布的彩色图像空域隐写检测算法[J].通信学报,2004,25(2):33-43.
[8]刘祖根,平玲娣,史烈,张宝军,潘雪增.基于图像子块颜色复杂度的隐写分析[J].浙江大学学报(工学版),2008,42(7):1150-1158.
[9]罗向阳,刘粉林,王道顺.基于小波包分解的图像信息隐写盲检测[J].通信学报,2008,29(10):173-183.
[10]陈丹,张键红,王育民.基于主成分分析的泛盲掩密分析方案[J].中国图象图形学报,2006,11(3):394-401.
[11]于禛,陈克非.基于图像的广义隐写分析[J].微电子学与计算机,2008,25(4):212-214.
[12]梁小萍,何军辉,李健乾,黄继武.隐写分析.原理、现状与展望[J].中山大学学报,2004,43(6):93-97.
[13]陈铭,张茹,钮心忻,杨义先.隐写分析技术研究概述[J].计算机应用,2008,28:31-33.
[14]A.Westfeld,A.Pfitzmann.Attacks on steganographic systems[C].Proceedings of Information Hiding,Third International Workshop,Dresden,Germany,1999,135-143.
[15]J.Fridrich and M.Goljan.Practical steganalysis:state of art[C].In SPIE security and watermarking of multimedia contents Ⅳ,San Jose,CA,2002,1-13.
[16]J.Fridrich,M.Goljan,D.Hogea.Steganalysis of JPEG images:breaking the F5algorithm[C].Workshop on 5~(th) information hiding,Noordwijkerhour,Netherlands,2002,310-323.
[17]Xiang Yang Luo,Dao Shun Wang,Ping Wang,Fen-Lin Liu.A review on blind detection for image steganography[J].Signal Processing,2008,88(9):2138-2157.
[18]Avicibas I,Memon N D,Sankur B.Steganalysis of watermarking techniques using image quality metrics[C].Proceedings of security and watermarking of multimedia contentsⅡ,San Jose,CA,USA,2001,523-531.
[19]Avicibas I,Memon N D,Sankur B.Steganalysis using image quality metrics[J].IEEE transaction on image processing,2003,12(2):221-229.
[20]姜楠,王健,杨义先.新的唯密载体信息隐写分析方法[J].北京邮电大学学报,2006,29(2):1-4.
[21]于禛.基于图像的广义隐写分析[D].中国优秀硕士学位论文全文数据库.上海交通大学:2008.
[22]H.Farid.Detecting steganographic messages in digital images[R].USA:Dartmouth College,2000.
[23]H.Farid.Detecting hidden message using higher-order statistical models[C].Proceedings of IEEE international conference on image processing,Rochester,New York,USA,2002,905-908.
[24]S.Lyu,H.Farid.Steganalysis using color wavelet statistics and one class support vector machines[C].SPIE Symposium on Electronic imaging,San Jose,CA,2004,35-45.
[25]S.Lyu,H.Farid.Steganalysis using higher-order image statistics[J].IEEE Transactions on Information and Forensics Security,2006,1(1):111-119.
[26]陈丹,陈原.一种针对彩色图像的泛盲掩密分析方案[J].电子与信息学报,2005,27(10):1542-1549.
[27]管超,周治平.改进的小波高阶统计量的JPEG图像隐写分析[J].计算机工程与应用,2008,44(25):193-195.
[28]Harmsen J,Pearlman A.Steganalysis of additive noise modelable information hiding [C].Proceedings of security,steganography and watermarking of multimedia contents,V.San Jose,CA,USA,2003,131-142.
[29]Xuan Guo Rong,Shi Yun Quan,Gao Jian Jiong,Zou Dekun,Yang Cheng yun.Steganalysis based on multiple features formed by statistical moments of wavelet characteristic functions[C].Proceedings of 7th International Information Hiding Workshop,Barcelona,Spain,2006.262-277.
[30]高建炯,宣国荣,施云庆,柴佩琪,杨志强.基于直方图频域统计矩的图像隐写分析[J].计算机应用,2005,25(10):2370-2374.
[31]黄聪,宣国荣,高建炯,施云庆.基于图像及其预测误差图小波频域矩的隐写分析[J].计算机应用,2006,26(8):1851-1854.
[32]黄聪,宣国荣,高建炯,施云庆.基于模式识别的多类隐写分析[J].计算机工程与应用,2006,(27):43-47.
[33]Wang Y,Moulin P.Optimized feature extraction for learning-based image steganalysis[J].IEEE Transactions on Information Forensics and Security,2007,2(1):31-45.
[34]Fridrich J.Feature based steganalysis for JPEG images and its implications for future design of steganographic schemes[C].Proceedings of 6th international workshop on information hiding,Roronto,Ontario,Canada,2004,67-81.
[35]Xiao Chuan Chen,Yun Hong Wang,Tieniu Tan,Lei Guo.Blind image steganalysis based on statistical analysis of empirical matrix[C].Proceedings of 18th international conference on pattern recognition,Hongkong,China,2006,1107-1110.
[36]罗向阳,王道顺,汪萍,刘粉林.基于图像多域特征缩放与BP网络的信息隐藏盲检测[J].东南大学学报,2007,37增刊(Ⅰ):87-94.
[37]Simmons G J.The history of subliminal channels[J].IEEE joumal on selected areas in communications,1998.16(4):452-462.
[38]吴婷.基于高频DCT系数的F5和outguess0.2图像隐写检测方法研究[D].中国优秀硕士学位论文全文数据库.上海交通大学:2006.
[39]Hsu C T,Wu J L.Hidden digital watermarks in images[J].IEEE transaction on image processing,1999,8:58-68.
[40]Provos N.Defending against statistical steganalysis[C].Proceedings of 10~(th) USENLX security symposium,Berkeley,CA,2001,323-335.
[41]Westfeld A.F5-A steganographic algorithm high capacity despite better steganalysis[C].Proceeding of 4~(th) international workshop on information hiding,Pittsburgh,USA,2001,289-302.
[42]何军辉.图像隐写分析研究[D].中国优秀博士学位论文全文数据库.中山大学:2006.
[43]杨淑莹.模式识别与智能计算—Matlab技术实现[M].北京:电子工业出版社,2008.
[44]孙延奎.小波分析及其应用[M].北京:机械工业出版社,2005.
[45]龚声蓉,刘纯平,王强等.数字图像处理与分析[M].北京:清华大学出版社,2006.
[46]高建炯.基于模式识别的图像隐写分析[D].中国优秀硕士学位论文全文数据库.同济大学:2006.
[47]I J Cox,J Kalian,T Leighton,T shamoon.Secure spread spectrum watermarking for multimedia[J].IEEE transaction on image processing,1997,6(12):673-687.
[48]Trahanias P E,Karakos D.Directional processing of color images:theory and experimental results[J].IEEE transaction on image processing,1996,5(5):868-880.
[49]Microsoft Research Cambridge Object Recognition Image Database[EB/OL].http://research.microsoft.com.
[50]Latham.Jpeg-jsteg[EB/CP].ftp://ftp.funet.fi/pub/crypt/steganography.
[51]Provos N.Outguess[EB/CP].www.outguess.org.
[52]Westfeld A.F5[EB/CP].http://wwwrn.inf.tu-dresden.de/～westfeld/f5.html.
[53]C-C.Chang,C.-Jilin.LIBSVM[EB/CP].http://www.csie.ntu.edu.tw/～cjlin/libsvm.
[54]Richard O.Duda,Peter E.Hart,David G Stork.模式识别[M].机械工业出版社,2006.
[55]刘乐平等.概率论与数理统计[M].江西高校出版社,2002,246-258.
[56]林关成,李亚安.基于ANN和SVM的分类和回归比较研究[C].促进中西部发展声学学术交流会论文集,拉萨,中国,2008,267-272.
[57]刘佳,杨晓元,吴艺杰.小波统计量和多类支持向量机的密写检测算法[J].计算机工程与应用,2006,7(26):105-109.
[58]赵春晖,陈万海,郭春燕.多类支持向量机的研究现状与分析[J].智能系统学报,2007,2(2):11-18.
