An edge-preserving filter algorithm of local average saliency gradient

To avoid the halo artifact in the traditional filters, a new edge-preserving filter based on local average saliency gradient was proposed. The criterion to determine whether or not a region should be smoothed is difficult to give, hence it will introduce the "halo" artifact along the edges by unwanted smoothing in an image. Our idea is motivate by the feature of saliency map that enhances the salient edges in an image. Using the feature, we simplify the work to determine whether or not a region should be smoothed, and smooth the image adaptively according to local average saliency gradient. We retain the region with high average saliency gradient, and smooth the region with low average saliency gradient. Experiments show that our method performs better than traditional edge-preserving filters in smoothing, multi-enhancement, and HDR while avoiding "halo".