Fusion of Multi-class Steg- Analysis Systems Using Bayesian Model Averaging

Several steganography methods are available over the Internet for hiding information within digital images. A digital forensics examiner must be able to extract a hidden message from a digital image. Extraction requires first identifying the embedding method used. Several steganalysis systems exist that identify a subset of the available embedding methods. Each steganalysis systems has its own particular identification strengths and weaknesses. This paper applies Bayesian model averaging to fuse these multiple systems and identify the embedding method used to create a stego JPEG image. The fusion of the multi-class detection systems results in better classification than the individual systems, with an accuracy improved from 80% for the individual multi-class steganalysis systems to 90% with the steganalysis fusion system.