Information embedding (IE) or watermarking is the transmission of a
message embedded into a host signal, in a way that is invisible to the
unintended observer. This problem has received considerable attention in recent
years, due to its wide range of applications, e.g., copyright protection
systems, source tracing and covert communications. The signal resulting of the
embedding process is called the composite signal, or the stegotext. It is
transmitted through a channel to the destination that decodes the message.
However, the user at the destination may be interested in retrieving the host
signal too. The quality of restoration of the host data can even be of utmost
importance in some applications such as medical imagery and military
communications. Reversible information embedding (RIE) extends the IE model by
adding the requirement of decoding the host signal. In some applications
however, this requirement is too strong and a high price has to be paid in the
rate of communication. For example, if the user is interested in decoding the
message and further transmitting it to another destination embedded into the
same host, it suffices to reproduce the stegotext at the destination. This work
expands the study of IE channels by adding the requirement of retrieving the
stegotext at the destination. This new model is called the Stegotext Reversible
Information Embedding (SRIE). A single-letter characterization of the
achievable rate-distortion region is developed. In particular, it is shown that
a rate higher than the RIE capacity can be achieved, under the (more relaxed)
requirement of stegotext decoding. The duality of this problem to the Common
Reconstruction model developed by Steinberg is discussed. Two binary examples
are solved, and an iterative algorithm that computes numerically the capacity
is provided. Then, an extension to the Multiple Access Channel for the SRIE
model is provided where two independent data streams are embedded into
dependent host signals. Inner and outer bounds on the capacity region are
developed. Finally, an extension of this model to the Broadcast Channel is
provided and a single-letter characterization of the capacity region is
derived.
