Unsharp Mask Guided Filtering

The goal of this paper is guided image filtering, which emphasizes the importance structure transfer during filtering by means an additional guidance image. Where classical filters structures using hand-designed functions, recent have been considerably advanced through parametric learning deep networks. state-of-the-art leverages networks to estimate two core coefficients filter. In work, we posit that simultaneously estimating both suboptimal, resulting in halo artifacts and inconsistencies. Inspired unsharp masking, a technique for edge enhancement requires only single coefficient, propose new simplified formulation Our enjoys prior from low-pass filter enables explicit coefficient. Based on our proposed formulation, introduce successive network, provides multiple results allowing trade-off between accuracy efficiency. Extensive ablations, comparisons analysis show effectiveness efficiency across tasks like upsampling, denoising, cross-modality filtering. Code available at \url{https://github.com/shizenglin/Unsharp-Mask-Guided-Filtering}.