Steganalysis: Trends and Challenges

1. Introduction

Steganography deals with hiding information into a cover (host or original) signal such that no one other than the intended recipient can detect or extract the hidden message. The steganographic encoder embeds a message into the cover-signal using a secret key such that perceptual and other distortion constraints are satisfied. A statistical dissimilarity measure between the cover and the stego-signal is generally used to measure the security of a given steganographic method (Cachin, 1998; Zollner et al., 1998; Chandramouli & Memon, 2003).

Steganography can be modeled as a prisoner’s problem (Simmons, 1984). For example, consider two prisoners, Alice and Bob, who want to secretly exchange information regarding their escape plan. However, the warden, Wendy, examines every communication between Alice and Bob, and punishes them if steganographic covert communication is detected. In a standard steganographic framework, Alice sends a secrete message, M, to Bob by embedding her secret message into the cover-signal, S, to obtain the stego-signal X. Alice then sends X to Bob using a public channel. The warden, who examines the communication channel between Alice and Bob, can be passive or active. A passive warden attempts only to detect a steganographic covert channel. An active warden, on the other hand, deliberately alters every signal exchanged between Alice and Bob, to foil any covert communication between them. The allowable distortion the warden can introduce in the stego-signal depends on the underlying model and the cover-signal used. Figure 1 illustrates secret key steganography for active and passive warden scenarios.

Figure 1

Secret key steganography in the presence of a passive warden (top) and an active warden (bottom)

Clearly, Alice and Bob attempt to design the steganographic channel (encoder, secret key, and decoder) such that the warden is unable to distinguish in any sense (statistically as well as perceptually) between the cover-signal and the stego-signal. On the other hand, Wendy tries to detect or estimating the hidden message, M, from the stego-signal, X by using one or several steganalysis algorithms. In general, steganalysis must not make any assumption about the underlying steganographic algorithm used to embed a secret message (Chandramouli, 2002).

Rapid proliferation of digital multimedia signals on the Internet makes them good candidates as cover-signals. The simplest of the existing steganographic techniques is the east significant bit (LSB) steganography. Steganographic techniques based on LSB embedding exploit the fact that in general digital images/video are perceptually insensitive to the distortion introduced in the least significant bit plane. These techniques embed the secret message, M, in the cover-signal by replacing the LSB plane of the cover-signal in the spatial/time domain or in the transformed domain. Table 1 illustrates LSB embedding with an example. Here, an 8-bit secret message “01010011” is embedded into eight samples of the cover-signal.