Abstract: An image diffusion algorithm based on visual masking effect is proposed. In view of properties of human visual system, inhomogeneity underlying an image is employed to define a novel noise masking function. It is a measure of contextual discontinuities, and could efficiently distinguish noise from edges. The resultant noise visibility function is used as diffusivity function of anisotropic diffusion. Due to the combined use of local spatial gradient and contextual information of larger neighborhood, the proposed algorithm could preserve nontrivial features more efficiently while removing noise. Further, it avoids the difficulty of estimating the gradient threshold in traditional algorithms. Comparative experiments certify that the proposed algorithm outperforms several existing typical diffusion methods in terms of noise removal and feature preservation.