Most watermarking techniques rely on redundancy where signature bits are en
coded in a sufficiently large amount of sites for robustness against distor
tions and attacks while the watermark energy is kept low enough to remain i
mperceptible. The goal of this paper is to explore some strategies for expl
oiting this redundancy using error correcting codes. In some watermarking t
echniques bits are extracted via hard-decisions leading to a binary symmetr
ic channel model while in others the extraction is carried out with soft-de
cisions leading to a additive Gaussian channel model. First, we consider er
ror correcting codes for very high error rates of the watermarking channel
where the trade-offs of Bose-Chaudury-Hocquenheim (BCH) and repetition code
s are investigated. We also present the performance and a fast implementati
on of soft-decoders. We give two potential realizations of soft-decoding, n
amely, Viterbi decoder for convolutional codes and a new algorithm for soft
-BCH decoding. (C) 2001 Elsevier Science B.V. All rights reserved.