A fast and accurate steganalysis using ensemble classifiers

Nowadays, the steganographic methods use themore sophisticated image models to increase security;consequently, steganalysis algorithm should build the moreaccurate models of images to detect them. So, the number ofextracted feature is increasing. Most modern steganalysisalgorithms train a supervised classifier on the feature vectors.The most popular and accurate one is SVM, but the high trainingtime of SVM inhibits the development of steganalysis. To solvethis problem, in this paper we propose a fast and accuratesteganalysis methods based on Ensemble classifier and Stacking.In this method, the relation between basic learners decisions andtrue decision is learned by another classifier. To do this, basiclearners decisions are mapped to space of uncorrelateddimensions. The complexity of this method is much lower thanthat of SVM, while our method improves detection accuracy.Proposed method is a fast and accurate classifier that can be usedas a part of any steganalysis algorithm. Performance of thismethod is demonstrated on two steganographic methods, namelynsF5 and Model Based Steganography. The performance ofproposed method is compared to that of Ensemble classifier.Experimental results show that the classification error andtraining time are lowered by 46% and 88%, respectively

Keywords:

Steganalysis , Ensemble classifiers , Correspondence analysis