基于半监督学习的JPEG图像通用隐写检测方法  

A Universal Steganalysis Method for JPEG Images Based on Semi-supervised Learning

作　　者：陈东泳[1] 钟尚平[1] 

机构地区：[1]福州大学数学与计算机科学学院,福建福州350002

出　　处：《计算机技术与发展》2009年第2期169-172,共4页Computer Technology and Development

基　　金：国家"863"高技术开发项目(2005AA142110);福建省青年人才基金项目(2006F3076)

摘　　要：目前,JPEG图像的通用隐写检测是基于监督学习的,其关键技术包括图像特征提取和分类器的设计。首次提出了运用半监督学习中的EM算法来进行分类器的设计,该方法利用大量未标记样本辅助少量有标记样本进行分类器的学习。针对经典的JPEG隐写方法:Outguess和F5,用监督学习与文中半监督学习方法进行实验对比,结果表明,在缺少大量标记样本的情况下,文中方法能得到较好的分类性能,从而提高了JPEG图像通用隐写检测方法的实用性。At present, the univemal steganalysis methods for JPEG images are -based on supervised learning, the key technologies of these methods indue image feature extraction and classifier design. Proposes a novel classifier whieh based on semi - supervised learning EM algorithm, the classifier makes use of a great quantity of unlabeled samples combined with a small quantity of labeled samples. Aimed at the popular JPEG steganography technologies： Outguess and F5, compared with this paper＇ s method and the supervised learning method, the experimental results show that the proposed method can obtain good performance when a large number of labeled samples can＇ t be obtained. Therefore, our method can improve the practicability for JPEG image universal steganalysis.

关 键 词：通用隐写检测 JPEG图像 半监督学习 EM算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]