基于模式识别的多类隐写分析

Multi-Class Steganalysis Based on Pattern Recognition

文章在小波频域高阶统计矩图象隐写分析基础上,将隐写分析问题扩充为多类问题,针对CorelDraw图象库5种常用的隐写方法进行了实验,结果表明该文方法不仅能检测出图象是否隐含数据,还能指出所用的隐写方法,识别率在90%以上。还提出了多类“类内类间分布图”,对于高维可分性好低维可分性差的图示问题,指导特征以及分类器的选择,具有很高的价值。

Based on our previous method,we propose Multi-Class Steganalysis,and the experiment using CorelDraw image database show that this method is not only able to tell whether images are stego or not,but also reliably classify stego images to their embedding techniques,the detection rate is above 90%.What＇s more,a method named Within-Class and Between-Class Distribution Graph（B-W Distribution Graph） is proposed,which can be used to evaluate the individual performance of every sample,provide information about the classifiability of features and hence assist in feature selection and classifier design.