基于GA和灰色SVC的图像通用密写分析

Universal Steganalysis for Image Based on Genetic Algorithm and Grey SVC

摘要: 在以支持向量分类机(Support Vector Classifier, SVC)为分类器的图像通用密写分析中，离群样本对最优分类面判别的影响成为导致分析性能不高的原因之一。文中提出一种新的密写分析算法。该算法首先在小波域上提取噪声信号的特征，然后利用遗传算法(Genetic Algorithm, GA)搜索最优类特征，最后将样本特征与最优类特征的灰色关联度参与到SVC的训练中，构造一个灰色支持向量机作为分类器。与采用相同特征的Holotyak算法相比，文中所提的通用密写分析算法的分类结果更好。

Abstract: The isolated samples can produce some affect on distinguishing the best classifying plane, which becomes one of causes of less performance of universal steganalysis that uses Support Vector Classfier (SVC) as classifier. This paper proposes a new universal steganalysis algorithm. The algorithm firstly catches characteristic of noise signal in wavelet domain. Then utilizes Gentic Algorithm (GA) search the best characteristic of species. Finally makes grey relational degree between sample characteristic and the best characteristic of species participate in training of SVC, thus constructs a Grey Support Vector Machines (GSVM) to be a classifier.