Quantitative Methods in Defense and National Security 2007

Fast and Efficient Steganalysis Methods for Spread Spectrum Steganography
Kai-Sheng Song, (Department of Statistics, Florida State University), kssong@stat.fsu.edu


Steganography, meaning ''covered writing'' in Greek, is the art and science of covert/hiding communication. The rise of the Internet and multimedia techniques has prompted increasing interest in hiding information in digital media including images, audio, video, and text. The goal of steganography, like invisible inks and the microdot that were widely-used in the past century, is to convey messages under cover, concealing the very existence of information exchange, which differs from cryptography that does not conceal the communication itself but only scrambles (camouflages) the data to prevent eavesdroppers from understanding the content. Many steganographic techniques have been proposed in the engineering and computer science literature such as spread spectrum image steganography, which is a data- hiding/hidden-communication method that uses digital imagery as a cover signal. Spread spectrum steganography is analogous to spread spectrum radio transmissions where the energy of the signal is spread across a wide-frequency spectrum rather than focused on a single frequency, in an effort to make detection and jamming of the signal harder. Spread spectrum steganography takes advantage of the fact that little distortions to digital signals such as image and sound files are least detectable in the high-energy portions of the carrier (i.e., high intensity in sound files or bright colors in image files).

Many data embedding schemes in the spread spectrum steganography literature assume that the noise and interference are i.i.d. Gaussian noise and thus the schemes can be modeled as communication over a channel with, for example, additive white Gaussian noise (AWGN). However, in practical applications, the noise and interference come from signal processing and/or attack in coherent detection where the original cover signal is available, and in the case of non-coherent detection, the noise and interference consist of the host media as well as signal processing and/or attack. Consequently, the white Gaussian assumption is very unrealistic and its limitations have been demonstrated by numerous experiments. For example, different bands of block discrete cosine transform (DCT) coefficients of many natural images have shown quite different variability.

In this paper, we present a statistical formulation of the spread spectrum steganography in the frame work of generalized Gaussian families with unknown parameters of shape and possibly different variances. Both additive and multiplicative schemes of spread spectrum steganography are investigated. Stego detectors in both coherent detection and non-coherent detection are derived from the efficient scores vector. Fast algorithms for implementing the procedures are also proposed. Furthermore, the asymptotic distributions of the detectors are established. Several numerical experiments using images as a vehicle for steganographic communication are conducted to demonstrate the performance of the proposed spread spectrum steganography methods.

Take me back to the main conference page.