Toward improved wavelet-based watermarking using the pixel-wise masking model

Barni's wavelet-based watermarking algorithm using a pixel-wise masking model shows great superiority over other methods in terms of the watermark imperceptibility and robustness against attacks including filtering, noise addition and compression. However, this algorithm adds the watermark into the three largest detail subbands, which gives the attackers the opportunity to remove it completely by just discarding these subbands. Moreover, Barni's pixel-masking model only uses the coarsest approximation subband to compute the local brightness and texture activity, which is usually too small to contain enough information. Following this pixel-wise masking idea, a new watermarking algorithm is presented in this paper, in which an improved version of the pixel-masking model that better describes the behavior of the HVS is designed, and a novel watermark embedder that adds the watermark into the most attack-resilient coefficients of all the detail subbands is proposed. Experiments verify that our algorithm is more robust than Barni's method.