Trellis Coded Quantization based watermarking scheme performances in an independent domain

In applications to secured diffusion of digital images, several watermarking solutions have been studied in the field of Data Hiding. For this aim, the main idea consists in embedding an information in the digital media with good properties in terms of robustness, capacity and invisibility (perceptual transparency and statistical undetectability). Recent studies has highlighted the importance of statistical undetectability which is unfortunately not often evaluated. We propose in this work to study a watermarking scheme with side-information at the encoder and based on Treillis Coded Quantization (TCQ). The main particularity of our approach is that we propose to apply the embedding process in an independent domain (by the way of Independent Component Analysis: ICA). We then show the improvement of the robustness and statistical undetectability properties with our proposal. The comparison with well known watermarking approaches, of the performances obtained, is done. This highlight the wide improvement of the obtained performances: high robustness without reducing the capacity, good security level or very low distorsions on the watermarked signals probability density functions.