Computer Engineering and Applications ›› 2007, Vol. 43 ›› Issue (23): 32-34.

• 学术探讨 • Previous Articles     Next Articles

JPEG steganalysis based on bidirectional Markov model

CUI Xia,TONG Xue-feng,XUAN Guo-rong,HUANG Cong,ZHU Xiu-ming   

  1. Department of Computer Science,Tongji University,Shanghai 200092,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2007-08-11 Published:2007-08-11
  • Contact: CUI Xia

基于双向马尔可夫模型的JPEG图象隐写分析

崔 霞,童学锋,宣国荣,黄 聪,朱秀明   

  1. 同济大学 计算机科学与工程系,上海 200092
  • 通讯作者: 崔 霞

Abstract: This paper presents a novel steganalysis scheme to attack JPEG steganography.The 600 dimensional feature vectors sensitive to data embedding process are derived from bidirectional Markov models in the DCT domain.The threshold Bayesian distance classifier is used to classify steganography in the high-dimensional feature vector space.In addition,SVM is used as a contrast.The experimental results have demonstrated that the proposed scheme outperforms the existing steganalysis technique in attacking modern JPEG steganographic schemes—F5,Outguess,MB1 and MB2.

Key words: steganalysis, JPEG, bidirectional Markov model, threshold Bayesian distance classifier, Support Vector Machine(SVM)

摘要: 提出了一种基于双向马尔可夫模型的JPEG图象的通用隐写分析方法,利用量化后分块DCT系数的中低频系数间的相关性,提取DCT块内和块间的特征,采用阈值贝叶斯分类方法进行识别,并且与SVM分类器的识别效果进行了比较。针对4种公认的JPEG嵌入方法——F5、Outguess、MB1和MB2进行隐写分析,在CorelDraw图象库上做实验,取得了很好的性能。

关键词: 隐写分析, JPEG图象, 双向马尔可夫模型, 阈值贝叶斯分类, SVM分类