Edit propagation for natural image using matting affinity

The popularity of stroke-based image/video appearance editing is increasing in image processing field because of the intuitive, easy-to-control interactive interface. In this paper, we propose a new approach to editing natural image that adopts a stroke-based workflow. In our framework, the user paints a few simple brush strokes on region of interest and specifies edit value on each region. These constraints are propagated to the rest of the image by building an energy minimization formulation under the policy that pixels with similar appearance should receive similar edits. In this paper, we design a new energy minimization formula which can be solved by a linear sparse matrix system. Rather than implying a global affinity of spatial locality and appearance differences in the formula, an efficient region-based matting affinity is used to measure the similarity of pixel pairs, which especially works well in editing fuzzy images. Experiment results show the effectiveness of our method, which is applicable for a variety of adjustments commonly performed in the digital darkroom.