Blind Detection Algorithm for BMP Stego Images Based on Feature Fusion and Ensemble Classification

Traditional blind detection techniques for BMP stego images mainly use a single feature set and a single classifier. However, a single feature set is difficult to completely reflect the differences caused by embedding, and a single classifier is also sensitive to samples. Therefore, we propose a blind detection algorithm based on feature fusion and ensemble classification to improve the accuracy of blind detection for BMP stego images. We firstly extract the features based on higher-order probability density function (PDF) moments of the decomposition subband coefficients and statistical moments of characteristic function (CF) of subband histograms, and then use serial feature fusion to construct a new feature set, adopt Bagging and RSM to train base classifiers and finally utilize the trained classifiers to detect images. The experiment results show that the proposed method can improve the accuracy of the common BMP steganographic methods, such as LSB replacement, LSB matching, SS, and QIM.