Pattern complexity-based JND estimation for quantization watermarking

Abstract

The perceptual just noticeable distortion (JND), which plays an important role in perceptual image watermarking and can refers to the optimal tradeoff between imperceptibility and robustness. Major challenges of the perceptual quantization watermarking approach are two-fold: (1) Most DCT-based JND models cannot accurately estimate the contrast masking effect due to the complicated interaction among visual contents. Research on cognitive science shows that the HVS is adaptive to extract the visual pattern for image understanding, and we can formulate the pattern complexity as another factor to determine the total watermarking strength. (2) Moreover, the calculated JND values will change as watermark embedding can affect the pixels of the image, i.e. the operation in cross-domain reduce the watermarking robustness. Therefore, in this regard, the maximum directional energy is calculated by three AC DCT coefficients, which can measure the different direction energy and keep the pattern complexity measurement insensitive to the changes caused by watermarking procedure. The luminance contrast is also calculated in the DCT domain. So a pattern complexity-based perceptual JND estimation model is designed by takeing DCT-based pattern complexity and luminance contrast into account. Furthermore, a new logarithmic quantization watermarking scheme is presented based on the proposed model to verify the feasibility and effectiveness of our proposed JND model. Experimental results show that the new built JND model can effectively enhance the robustness of the quantization watermarking scheme.