Research On Speech Steganalysis Based On Ensemble Learning

With the progress of informatization technology and the rapid development of internet, there are more and more attention to steganography and steganalysis. Steganography is a technology which embeds secret messages into carrier signal. The purpose of steganalysis is making statistical analysis of the multimedia signals and determining whether secret information has been hidden in. Generally, almost all researches use single classifier as learner. Ensemble learning is a new machine learning paradigm. It can significantly improve the generalization ability of learning systems through utilizing multiple learners to solve a problem.This paper uses ensemble learning, which will combine the weak classifiers to a strong classifier, to improve the accuracy of steganalysis. Echo information hiding is a commonly-used method of speech information hiding. Focusing on the analytical problem of speech echo hiding, a steganalysis method based on cepstrum ensemble learning is proposed. By analyzing the principle of the cepstrum and ensemble learning, the steganalysis system of echo based on AdaBoost ensemble learning algorithm is built. The features that contained cepstrum, the first-order and second-order cepstrum, the mel-cepstrum, combined histogram and moments of high-order statistics are simulated by using SVM and AdaBoost ensemble learning algorithm. Simulation results show that the classification effects of the ensemble learning are better than the SVM for the cepstrum, the first-order and second-order cepstrum and combined histogram features. In addition, a time-spread echo hiding steganalysis algorithm based on ensemble learning is also proposed, and the same characteristics are simulated by using SVM and AdaBoost ensemble learning algorithm of the simulation analysis. Experiment results further verify the better generalization ability and the higher detection accuracy of ensemble learning than the ordinary classifier.