摘要
针对目前大部分BMP隐写分析方法主要采用单一特征和单一强分类器,容易产生训练样本敏感、分类精度难以提高等问题,提出一种基于特征融合和改进RSM集成分类的BMP图像隐写检测方法.方法首先串行融合Moulin和SPAM两种经典特征,然后利用序列前向选择(SFS)算法选取分类能力高的特征作为固定特征,其余特征在剩余特征空间中随机抽取,利用固定特征和随机抽取特征构造特征子集,最后在特征子集上训练成员分类器,并用多数投票法对它们进行组合.实验结果表明:和传统方法相比,在不同嵌入率下,该方法对BMP经典隐写(如LSB匹配、LSB替换、SS和QIM)的检测率均有一定程度的提高.
At present,most of the method of BMP steganalysis mainly adopted the single feature and a single strong classifier,which prone to the problems of training samples sensitivity and difficult to improve the classification accuracy.In order to solve these problems,the method of BMP steganalysis is proposed based on feature fusion and improved RSM ensemble.The method firstly serial fusion Moulin feature and SPAM feature.Then selected features which have high classification ability using SFS algorithm as fixed features,the remaining features were selected randomly in the remaining feature space,and then the feature subset was build using fixed features and features selected randomly,finally member classifier was trained on the subset of features,and the finial decision was made by the majority voting procedure.Experimental results show that,in various embedding rates,this method provides more accuracy than the traditional method against steganographic methods of BMP （e.g.LSB matching,LSB replacement,SS and QIM）.
出处
《福州大学学报（自然科学版）》
CAS
CSCD
北大核心
2014年第5期661-665,677,共6页
Journal of Fuzhou University(Natural Science Edition)
基金
福建省教育厅科研资助项目(JB09003)
