基于四元数多特征并行融合的JPEG隐写检测

JPEG steganalysis based on parallel feature fusion by quaternion

针对串行特征融合方法易出现"维数灾难"以及并行复矢量特征融合方法只能融合两类特征的弱点,提出一种基于四元数多特征并行融合的JPEG隐写检测方法。方法利用四元数有4个分量能融合4种特征的性质,首先提取4种经典特征,然后用主成分分析(PCA)进行数据降维,去除冗余信息,最后将4种特征组合为四元数矢量,实现多特征的并行融合。实验结果表明,和传统特征融合方法相比,所提方法不仅有效提高了JPEG隐写图像检测率,而且具有较强的鲁棒性。

Feature fusion has become a heated research topic in steganalysis. In order to solve the problems in the existing feature fusion methods, such as feature dimension disaster in serial feature fusion method and only two kinds of features can be integrated in parallel feature fusion method, this paper proposed a JPEG steganalysis approach based on parallel feature fusion by quaternion. The method firstly selected the four fusion features, and then isolated redundant components by the Principal Component Analysis （PCA）, finally combined four features as quaternion vector to achieve multi- feature parallel fusion using the four components of the quaternion. The experimental results show that, compared with the traditional feature fusion method, the proposed method effectively improves the detection rate of steganalysis in JPEG images, and has better robustness.