Due to fears of secret messages being passed by terrorists, steganalysis has become topical. Hiding messages within a container or cover presents several problems, including the difficulty of detecting embedded information (existence and size of specific messages, or features within); the issue of restoring damaged containers; and the possibility of evading detection. Not all of these problems differ significantly from standard image analysis issues, and this emerges as a key to this paper’s intent—to find and fill a gap in the literature.

A general description of the problem is characterized as a function on sets x, m, and k: containers, messages, and secret keys (known to sender and receiver). Partial information on these is assumed to be available to would-be detectors, represented, for example, for m and m as a random value &thgr;(m, k). The author focuses on &thgr; when the stego-message y = f(x,m,k) is given, and x when y is given and &thgr; is known or evaluated. Identifying similarities to well-known hidden Markov model (HMM) solutions leads to standard solution methods that are then applied to an example of least significant bit (LSB) manipulation.

Sidorov’s simulation results provoke switching to random Markov field models to cover deficiencies. Container and encoding issues affecting appropriateness are not covered, but are implicit in the cited results. A claimed novelty lies in promoting established image analysis methods, particularly when containers are images; this point serves the author’s appeal for more HMM research in steganalysis. A claimed uniformity rests on “hiddenness” in the principal methods discussed, despite differences in solution attacks. (Markov chains are well represented in the field.)