面向JPEG图像的隐写分析盲检测方法  

A Blind Steganalytic Method to Detect JPEG Image Steganography

作　　者：王丽娜[1,2] 王汉森 翟黎明[1,2] 徐一波[1,2] 任延珍[1,2] WANG Lina;WANG Hansen;ZHAI Liming;XU Yibo;REN Yanzhen(Key Laboratory of Aerospace Information Security and Trust Computing, Ministry of Education, Wuhan University, Wuhan 430072, Hubei, China;School of Cyber Science and Engineering, Wuhan University, Wuhan 430072, Hubei, China)

机构地区：[1]武汉大学空天信息安全与可信计算教育部重点实验室,湖北武汉430072 [2]武汉大学国家网络安全学院,湖北武汉430072

出　　处：《武汉大学学报（理学版）》2018年第3期217-224,共8页Journal of Wuhan University:Natural Science Edition

基　　金：NSFC-通用技术基础研究联合基金(U1536204)

摘　　要：隐写分析盲检测存在着检测模型的检测准确性和通用性难以兼顾的问题.本文提出一种用于隐写分析的快速支持向量分类算法FC-SS2LM（fast classification for small sphere with two large margins）,通过构造最小超球体和双边最大间隔隐写分析模型,使检测模型既能准确构造分类边界又能考虑不同隐写样本的分布特点,达到了兼顾检测准确性和通用性的目的.在BOSSBase标准图像库上对提出的隐写分析盲检测模型进行验证,实验结果表明,该方法在一定程度上克服了传统隐写分析模型通用性差的缺点,同时提高了实际应用中训练数据样本不平衡情况下的检测准确率.即使在实际应用中训练集样本过大、支持向量较多的情况下,采用该方法计算也可以减小算法复杂度,提高泛化能力和分类速度.The current methods of blind steganalysis have difficulty in balancing the accuracy and versatility in detection.Therefore,a fast support vector classification algorithm FC-SS2 LM（fast classification for small with sphere two large margins）is proposed in this paper.By constructing a steganalysis model of small hypersphere with two large margins,we can not only construct the accurate classification boundary but also consider the distribution characteristics of different steganography samples,then both the accuracy and versatility in detection can be achieved.The experiments on the BOSSBase standard image dataset show that this method,to some extent,overcomes the shortcomings of the poor versatility of traditional steganalysis models and the imbalance of data samples in practical applications.Meanwhile,it also reduces the complexity of algorithm,improves the generalization ability and increases the speed of classification especially in the case of large number of support vectors.

关 键 词：盲检测 隐写分析 快速分类 通用隐写分析 

分 类 号：TP309.2[自动化与计算机技术—计算机系统结构]