Adaptive Energy Diffusion for Blind Inverse Halftoning

We propose a blind inverse halftoning method with adaptive energy diffusion. A discrete Voronoi diagram is built by treating halftone clots as Voronoi cell sites. Gaussian filters are then created adaptively based on Voronoi cells and used for energy diffusion to form the grayscale inverse halftone image. We further perform a median filter on the Gaussian filters' parameters to maintain consistency of filters across different image regions. To remove artifacts in dense halftone dots area, we show that a secondary Voronoi diagram can be built based on the non-halftone dots and a heuristic blending can be employed. Comparing with other inverse halftoning methods, our method is able to remove artifacts in both sparse and dense halftone dots regions and produces visually pleasant gray-scale images. Our method is also applicable to edge enhancement and structure-aware halftoning.