Physical Sciences and Mathematics Commons^™

Multi-Class Classification Averaging Fusion For Detecting Steganography, Benjamin M. Rodriguez, Gilbert L. Peterson, Sos S. Agaian

Faculty Publications

Multiple classifier fusion has the capability of increasing classification accuracy over individual classifier systems. This paper focuses on the development of a multi-class classification fusion based on weighted averaging of posterior class probabilities. This fusion system is applied to the steganography fingerprint domain, in which the classifier identifies the statistical patterns in an image which distinguish one steganography algorithm from another. Specifically we focus on algorithms in which jpeg images provide the cover in order to communicate covertly. The embedding methods targeted are F5, JSteg, Model Based, OutGuess, and StegHide. The developed multi-class steganalvsis system consists of three levels: (1) …