Abstract: Focused on the property of error diffuse halftoning, a multi-scale pyramid transform for two dimensions is constructed by the nonlinear median operator. Using the transform distribution of halftone noise in approximations and details on each-scale and its de-noising method are studied through many experiments, and a novel inverse halftoning algorithm via median pyramid transform is proposed. The method firstly decomposes the halftoning image by the transform in one-scale, and a wiener filter is applied to the detail coefficients. Then an inverse image is reconstructed from the cleaned details and approximations. Experimental results demonstrate that the proposed transform can separate the halftone noise and details better than the linear multi-scale transform. The proposed algorithm has the efficient edge preserving ability while smoothing the halftone patterns, excellent PSNR performance, low time complexity and small memory requirement.

Key words: halftoning, inverse halftoning, median pyramid transform, probability distribution function