A Study of Content based Watermarking using an Advanced HVS Model

A key point of image watermarking schemes is to select the image pixels or transformed coefficients where an embedded watermark would reach the optimal invisibility versus robustness trade-off. Human visual system (HVS) based JND masks are thus useful to optimally adapt the watermark strength right below the visibility threshold. Most of the perceptual masks created so far for watermarking purpose are based on the empirical assumption that edges and/or textures allow the best watermark visual masking and thus ensures good invisibility properties. Nevertheless, although invisibility control in smooth areas is challenging, it could be interesting to extend the watermark in such areas since the robustness performances might increase. The goal of this paper is to use an advanced HVS model to manage invisibility in order to evaluate the robustness of watermarks embedded in smooth areas only, edges/textured areas only or in both areas. This study is further to the one presented in (F. Autrusseau and P. Le Callet, 2007). The robustness was tested against 89 attacked images (using Stirmark), results showed that including smooth areas in the JND mask increases the robustness while preserving good invisibility properties.