Towards Optimal Rate Allocation For Logo Watermarking Using Reversible DWT

We introduce a novel logo watermarking algorithm using reversible discrete wavelet transform. With one level less decomposition than the host image, the logo coefficients of each subband are embedded into the corresponding subbands of the host image without using the approximate subband (LL) based on a well-known model of human vision system (HVS). Rather than manually tuned global scaling factors to control the embedding strength, we propose to assign different scaling factors for different subbands. Under this scheme, the logo watermarking could be viewed as a logo-compression-and-watermarking problem, in which logo coefficients are first quantized by different step-sizes (the inverse of scaling factors) based on both the perceptual weights and the subband statistical information, and then embedded. The step-sizes are optimized by a rate allocation algorithm. Equivalently this is a quality-constraint watermarking, where the embedding energy is adaptively distributed throughout different subbands of the host image. Numerical results verify the effectiveness of the system and the robustness of the algorithm against various types of general attacks. Reversible DWT makes the system quite suitable for hardware implementations