KCCA feature fusion in universal steganographic detection

Abstract

Feature fusion method has improved steganographic detection performance based on classical feature, however there are some drawbacks of this: without analysing the correlation of the basic features, it's only a simple combination of features and lacks standard for features selection; serial fusion feature always has high dimension, which will lead great time cost and possibility of “curse of dimensionality”. In this paper, we proposed a novel framework for measuring the feature selection and fusing two selected feature sets in steganographic detection field, based on KCCA theory. KCCA feature fusion method can outperform single feature and achieve similar performance to serial feature fusion method in steganographic detection field, while only costing 1/10∼1/8 of original time. So it has better practicability.