基于微观模板的多类JPEG图像盲隐密分析方法

Multi-class Blind Steganalysis for JPEG Images Based on Micro-templates

为准确判断一幅JPEG图像使用了何种隐密软件,针对JPEG隐密软件可能采用的DCT域隐密操作,建立了基于微观模板的统计特征空间,并在此基础上提出了一种多类JPEG图像盲隐密分析方法。实验结果表明:对JSteg、F5和Outguess 3种典型JPEG图像隐密软件各自生成的、含密量大于20%的JPEG隐密图像,该方法的隐密软件识别正确率均在97%以上。基于微观模板的统计特征提取方案可有效区分不同的DCT域隐密操作,有助于对JPEG隐密软件的检测判决。

To judge which steganographic algorithm is used for a JPEG image, some possible steganographic operations in DCT domain is investigated and a new steganographic feature space is constructed based on the methodology of micro-templates. Supported by this feature space, a new multiclass blind steganalyzer for JPEG images is presented. Experimental results indicate that, for stegoimages with secret message length greater than 20% , which are produced by typical steganographic software, such as JSteg, F5 and Outguess respectively, the steganalyzer can reliably classify all the stego-images to their embedding techniques with high accuracy above 97%. The methodology of micro-templates can effectively distinguish different steganographic operations in DCT domain, which favors the detection of JPEG steganographic algorithms.