TOWARDS ENERGY OPTIMIZATION OF TRANSPARENT WATERMARKS USING ENTROPY MASKING

摘要

An invisible image watermarking scheme tends to insert the mark in a transparent manner, with the highest possible energy. Several watermarking algorithms use human visual models as perceptual masks to maximize the power of watermark. We take advantage of inhibitory and excitatory characteristics of eye receptors which yield an extra capacity, for amplifying watermark power, while still satisfying imperceptibility constraint. Spatial-domain image entropy is used as a measure of strength of inhibitory effect. The principle of our idea is based on the decomposition of mark sequence into two parts, corresponding to two complementary sets of pixels in spatial-domain. These sets are determined with respect to two non-overlapping entropy intervals. We introduce entropy masking as a post-amplification process, which enables us to achieve exact localization. We derive formulas to refine Watson's visual model in a completely different manner, with the assumption of having a priori information about mark sequence. Simulation results of the proposed model confirm the expected both energy improvement and recovery enhancement of the watermark.