A Classifier Ensemble Algorithm Based on Improved RSM for High Dimensional Steganalysis

Today, ensemble learning algorithms are proposed to address the challenges of high dimensional classification for steganalysis caused by the curse of dimensionality and obtain superior performance. In this paper, we propose a classifier ensemble algorithm based on improved Random Subspace Method (RSM) for high-dimensional blind steganalysis. Firstly, sequential forward selection (SFS) algorithm is adopted to select part of features with high classification ability as fixed subset, so that the original feature space is partitioned into two parts: fixed subset and remaining feature subset, then the final feature subset is formed by selecting features randomly in each part according to the given sampling rate. Secondly, each base classifier is trained on the feature subset and the weight of each classifier is computed according to the classification accuracy and mutual information. Finally, the final decision is yielded using the weighted voting. Experiments with the steganographic algorithms HUGO demonstrate that the proposed algorithm can effectively increase the classification accuracy, in most cases, the detection accuracy is better than RSM and other classical classifier ensemble methods.