J4 ›› 2015, Vol. 42 ›› Issue (5): 133-138.doi: 10.3969/j.issn.1001-2400.2015.05.023

• 研究论文 • 上一篇    下一篇



  1. (1. 武警工程大学 电子技术系,陕西 西安  710086;
    2. 密码与信息安全保密武警部队重点实验室,陕西 西安  710086;
    3. 武警福州指挥学院,福建 福州  350002)
  • 收稿日期:2014-05-10 出版日期:2015-10-20 发布日期:2015-12-03
  • 通讯作者: 张敏情
  • 作者简介:张敏情(1967-), 女, 教授, E-mail:api_zmq@126.com.
  • 基金资助:


Blind steganalysis algorithm based new calibration for JPEG images

ZHANG Minqing1,2;ZHANG Yan3;LI Delong1;LUO Peng1,2   

  1. (1. Dept. of Electronic Technology, Engineering Univ. of the CAPF, Xi'an  710086, China;
    2. Key Lab. of CAPF for Cryptology and Information Security, Xi'an  710086, China;
    3. Fuzhou Command Academy of the Chinese Armed Police Force, Fuzhou  350002, China)
  • Received:2014-05-10 Online:2015-10-20 Published:2015-12-03
  • Contact: ZHANG Minqing


为进一步提高隐写分析中校准特征对嵌入的敏感性,通过分析校准技术及其与特征之间的关系,在校准已有的分类基础上建立了一种数学模型,提出了一种基于该校准特性的JPEG通用隐写分析算法.算法采用剪切4像素的校准操作,结合微分知识提出校准的改进形式,根据校准前后图像特征的空间分布,得到直方图特征; 再由冗余关系计算新校准表示下的马尔可夫转移概率矩阵,最后与块间特征融合后得到新特征集.通过对nsF5、Jsteg和MB1算法在较低嵌入率时的检测和新特征集子集比较实验,发现该方法较现有校准分析方法具有更好的检测性能,达到了90%以上的正确率; 各特征集也表现了一定的互补性;在不同质量因子实验中性能较为稳定,可靠性较好.

关键词: 校准, JPEG隐写分析, 特征融合, 可靠性


To improve the embedding sensibility of calibrated feature in steganalysis, by studying the relationship between calibration technique and feature, a mathematical model for calibration based on the calibration classification is established, and a blind JPEG steganalysis algorithm based on the new calibration is presented. First we crop 4 pixels in the image and put forward a modified form of calibration, then the histogram characteristic is obtained according to the spatial distribution of the image features before and after calibration, and the Markov transfer probability matrix of the new calibration is calculated on the basis of redundancy. Finally, we fuse these features with the blocks feature and obtain the feature vector.Through the detection experiment of nsF5, Jsteg and MB1 algorithms with low embedding rates and among the feature vector, it is shown that this method has a better detection performance compared with those existing calibration methods. Its correctrate is more than 90%. The feature sets also show some complementary characteristics. It can be more stable and reliable in the different quality factors experiment.

Key words: calibration, JPEG steganalysis, feature fusion, reliability


  • TP391