Salt and pepper noise removal by adaptive median filter and minimal surface inpainting

In this paper, we propose a two-stage method for denoising images heavily contaminated by salt and pepper noise. In the first stage, we use adaptive median filter to detect the noisy pixels. All the pixels are marked as noisy or noisefree pixels. In the second stage, we take the noisy pixels as inpainting regions and noise-free pixels as true information. The inpainting task is done by the normalized mean curvature flow. The method is generalized to color image. Experimental results show that the proposed algorithm has advantages over nonlinear filtering or regularizing methods in terms of edge preservation and noise removal and is competitive with other two-stage methods in the literature.