JPEG Steganalysis Based on Co-occurrence Features and Ensemble Multiple Hyperspheres OC-SVM

Steganography is the technology of hiding a secret message in plain sight. The goal of steganalysis is to detect the presence of embedded data and to eventually extract the secret message. Current blind steganalytic methods, which relied on two-class or multi-class classifier, have offered strong detection capabilities against known embedding algorithms, but they suffer from an inability to detect previously unknown forms of steganography. In this paper, a new JPEG blind steganalytic technique for detecting both known and unknown steganography is proposed. On the basis of co-occurrence features and multiple hyperspheres One-Class SVM(OC-SVM) classifier, the proposed method can effectively model the statistics distribution boundary of innocent JPEG images. Bagging ensemble learning algorithm is also used to achieve higher detecting performance. Experimental results show the superiority of the method over other analogous steganalytic techniques.