MODEL-BASED METHODS FOR STEGANOGRAPHY AND STEGANALYSIS

Abstract

This paper presents methods for performing steganography and steganalysis using a statistical model of the cover medium. The methodology is general, and can be applied to virtually any type of media. It provides answers for some fundamental questions that have not been fully addressed by previous steganographic methods, such as how large a message can be hidden without risking detection by certain statistical methods, and how to achieve this maximum capacity. Current steganographic methods have been shown to be insecure against simple statistical attacks. Using the model-based methodology, an example steganography method is proposed for JPEG images that achieves a higher embedding efficiency and message capacity than previous methods while remaining secure against first order statistical attacks. A method is also described for defending against "blockiness" steganalysis attacks. Finally, a model-based steganalysis method is presented for estimating the length of messages hidden with Jsteg in JPEG images.