基于信噪比和相关信息熵的彩色图像隐写检测特征选择方法  

Feature Selection Method of Color Image Steganalysis Based on Signal-to-Noise Ratio and Correlation Information Entropy

作　　者：徐久成 杨杰 孟祥茹[1,2] 瞿康林 马媛媛 XU Jiucheng;YANG Jie;MENG Xiangru;QU Kanglin;MA Yuanyuan(College of Computer and Information Engineering,Henan Normal University,Xinxiang,Henan 453007,China;Engineering Lab of Intelligence Business&Internet of Things,Henan Normal University,Xinxiang,Henan 453007,China)

机构地区：[1]河南师范大学计算机与信息工程学院,河南新乡453007 [2]智慧商务与物联网技术河南省工程实验室,河南师范大学,河南新乡453007

出　　处：《昆明理工大学学报（自然科学版）》2023年第1期31-41,共11页Journal of Kunming University of Science and Technology(Natural Science)

基　　金：国家自然科学基金项目(61976082,62002103)。

摘　　要：彩色图像隐写检测方法在提取特征的时候带来大量的冗余特征,降低了检测精度.在特征选择中,信息论方法是将特征相关性和冗余性进行分割判断,无法判断整个特征子集的组合效应.因此,将相关信息熵引入到彩色图像隐写检测特征选择中,提出了一种基于信噪比和相关信息熵的彩色图像隐写检测特征选择方法.首先,使用信噪比度量彩色图像的每一维隐写检测特征对样本分类的贡献度,去除不相关特征获得预处理子集;然后,利用相关信息熵度量彩色图像隐写检测特征之间的独立性和冗余程度,设计了一种特征排序算法对预处理子集进行特征重要度排序,使用集成分类器剔除冗余特征,以得到最终约简子集.实验结果表明,所提方法能在不影响检测精度的同时有效降低特征维数,且优于C-FNCES、CGSM和Steganalysis-α隐写检测特征选择方法.The color image steganalysis method brings a large number of redundant features while extracting features,which reduces the detection accuracy.In feature selection,the traditional information theory method is to segment and judge the feature correlation and redundancy,which can not assess the combination effect of the whole feature subset.In this paper,the correlation information entropy is introduced into the feature selection of color image steganalysis,and a feature selection method of color image steganalysis based on signal-to-noise ratio and correlation information entropy is proposed.Firstly,the signal-to-noise ratio is used to measure the contribution of each one-dimensional steganalysis feature to sample classification,and the irrelevant features are removed to obtain the preprocessed subset.Then,a feature ranking algorithm is designed to obtain the final reduced subset by using the correlation information entropy to measure the independence and redundancy between the color image steganalysis features.The experimental results show that the proposed method can effectively reduce the feature dimension while maintaining or even improving the detection accuracy,and is better than C-FNCES,CGSM and Steganalysis-αmethod.

关 键 词：信噪比 相关信息熵 特征选择 彩色图像 隐写检测 

分 类 号：TP309.7[自动化与计算机技术—计算机系统结构]