Unified Spatial Masking for Just-Noticeable Difference Estimation

In this paper, we introduce a unified spatial masking function for the estimation of just-noticeable difference (JND). Conventional models estimate several parts independently, and then combine these parts to get the JND. In this work, we treat the spatial masking effect as a nonlinear transformation of the luminance adaptation. To model the transformation, we measure the deviation of image contents from the ideal patterns to establish luminance adaptation rules. Considering both luminance difference and structural regularity, we derive a nonlinear spatial masking function by modulating luminance adaptation with the deviation coefficients. The masking function deduces an accurate estimation of the JND. Experiments demonstrate the validity of the proposed framework.