[1]江夏秋,王丽娜,胡东辉,等.一种基于无监督学习的MB1隐写分析方法[J].东南大学学报(自然科学版),2009,39(3):442-446.[doi:10.3969/j.issn.1001-0505.2009.03.005]
 Jiang Xiaqiu,Wang Lina,Hu Donghui,et al.MB1 steganalysis based on unsupervised learning method[J].Journal of Southeast University (Natural Science Edition),2009,39(3):442-446.[doi:10.3969/j.issn.1001-0505.2009.03.005]
点击复制

一种基于无监督学习的MB1隐写分析方法()
分享到:

《东南大学学报(自然科学版)》[ISSN:1001-0505/CN:32-1178/N]

卷:
39
期数:
2009年第3期
页码:
442-446
栏目:
计算机科学与工程
出版日期:
2009-05-20

文章信息/Info

Title:
MB1 steganalysis based on unsupervised learning method
作者:
江夏秋1 王丽娜12 胡东辉13 岳云涛1
1 武汉大学计算机学院, 武汉430079; 2 空天信息安全及可信计算教育部重点实验室, 武汉 430072; 3 合肥工业大学计算机与信息学院, 合肥 230009
Author(s):
Jiang Xiaqiu1 Wang Lina12 Hu Donghui13 Yue Yuntao1
1 School of Computer Science, Wuhan University, Wuhan 430079, China
2 Key Laboratory of Aerospace Information Security and Trust Computing, Ministry of Education, Wuhan 430072, China
3 School of Computer Science and Information, Hefei University of Technology, Hefei 230009, China
关键词:
隐写分析 无监督学习 MB1隐写算法 支持向量数据描述法
Keywords:
steganalysis unsupervised-learning MB1 steganography support vector domain description
分类号:
TP391
DOI:
10.3969/j.issn.1001-0505.2009.03.005
摘要:
为了提高针对混杂小样本集的MB1隐写算法的检测率,提出了一种泛化能力较强的MB1隐写分析方法.通过分析多种图像特征,在离散余弦变化(DCT)域选取对隐写敏感易变的特征,包括变分特征、块边界度量特征、共生矩阵特征和马尔可夫特征组成的108维特征向量,并以无监督学习中的支持向量数据描述法(SVDD)为分类器,使用含有混杂样本的小样本集进行训练,测试算法对隐写图像的检测率.实验结果表明,当检测相对嵌入率为40%以上的隐写图像时,检测率可靠度达到96%以上,明显高于其他2种基于支持向量机的经典算法.这说明本方法打破了其他方法对训练样本集的限制,提高了对混杂小样本集的MB1隐写算法的检测率.但由于它对混杂样本具有一定的容忍度,对较小嵌入率的隐写图像的检测率稍低.
Abstract:
To improve the detection rate of MB1 steganography based on noisy small sample sets, a steganalysis method with high generalizability is proposed. By analyzing a variety of image features, the features in the Discrete Cosine Transform(DCT)domain which are sensitive to the steganography are selected, including variation, blockiness, co-occurrence matrix and Markov characteristics. These features consist of 108-dimensional feature vectors. The support vector data description is used as a classifier. And a small sample set containing unpurified samples is used for training. Test results show that when the relative embedding rate of the stego-images reaches 40%, the detection reliability of this method is above 96%, which is significantly higher than the two state-of-the-art algorithms based on support vector machine. It reveals that this method overcomes the restriction of training sample sets demanded by other approaches, and improves the detection rate of MB1 steganography based on noisy small sample sets. However, due to the tolerance of mixed samples only within a certain degree, the detection rate for stego images with low embedding rates is slightly low.

参考文献/References:

[1] Fridrich J.Feature-based steganalysis for JPEG images and its implications for future design of steganographic schemes[C] //The 6th Information Hiding Workshop.Berlin:Springer-Verlag,2004:67-81.
[2] Pevny T,Fridrich J.Merging Markov and DCT features for multi-class JPEG steganalysis[C] //Proc SPIE Electronic Imaging,Security,Steganography and Watermarking of Multimesdia Contents IX.San Jose,USA,2007:650503-650504.
[3] Shi Y Q,Chen C,Chen W.A Markov process based approach to effective attacking JPEG steganography[C] //Proceedings of the 8th Information Hiding Workshop.Berlin:Springer-Verlag,2006:849-852.
[4] 黄聪,宣国荣,高建炯,等.基于DCT域共生矩阵的JPEG图像隐写分析[J].计算机应用,2006,26(12):2863-2865.
  Huang Cong,Xuan Guorong,Gao Jianjiong,et al.Steganalysis based on co-occurrence matrix in DCT domain for JPEG images[J].Journal of Computer Applications,2006,26(12):2863-2865.(in Chinese)
[5] 崔霞,童学锋,黄聪.基于马尔可夫模型和支持向量机的JPEG图像隐写分析[J].计算机应用,2007,27(9):2140-2142.
  Cui Xia,Tong Xuefeng,Huang Cong.Steganalysis based on Markov model and SVM for JPEG images[J].Journal of Computer Applications,2007,27(9):2140-2142.(in Chinese)
[6] 崔霞,童学锋,宣国荣,等.基于双向马尔可夫模型的JPEG图象隐写分析[J].计算机工程与应用,2007,43(23):32-34.
  Cui Xia,Tong Xuefeng,Xuan Guorong,et al.JPEG steganalysis based on bidirectional Markov model[J].Computer Engineering and Applications,2007,43(23):32-34.(in Chinese)
[7] Tax David M J,Duin Robert P W.Support vector domain description[J].Pattern Recognition Letters,1999,20(11):1191-1199.
[8] Greenspun Philip.Greenspun image library [EB/OL].(2004-01-02)[2008-06-10].http://philip.greenspun.com/photography.
[9] Tax David M J.Data description toolbox [EB/OL].(2003-09-26)[2008-05-15].http://ida.first.fhg.de/~davidt/dd_tools.html.
[10] Chang Chih Chung,Lin Chih Jen.LIBSVM:a library for support vector machines [EB/OL].(2008-04-08)[2008-09-20].http://www.csie.ntu.edu.tw/~cjlin/libsvm.

备注/Memo

备注/Memo:
作者简介: 江夏秋(1986—),女,硕士生; 王丽娜(联系人), 女, 博士, 教授, 博士生导师,lnawang@163.com.
基金项目: 国家自然科学基金重大研究计划资助项目(90718006)、国家自然科学基金资助项目(60743003)、教育部科学技术研究重点项目资助项目(108087)、国家教育部博士点基金资助项目(20070486107).
引文格式: 江夏秋,王丽娜,胡东辉,等.一种基于无监督学习的MB1隐写分析方法[J].东南大学学报:自然科学版,2009,39(3):442-446. [doi:10.3969/j.issn.1001-0505.2009.03.005]
更新日期/Last Update: 2009-05-20