A robust denoising method for random-valued impulse noise based on smooth and texture region separation
Zhenyu Liu, Yi Wan · 2012
So far it is still difficult to remove random-valued impulse noise excellently at any noise level for any image. In this paper, we present a novel robust two-phase denoising scheme based on smooth and texture region separation. In the first phase, all pixels are classified into two groups by using cascade window filtering: the smooth region pixels and the texture region pixels. Then we identify pixels which are likely to be contaminated by noise in these two groups separately. In the second phase, detail-preserving regularization [13] is used to restore the image. Extensive simulations demonstrate the robustness of this proposed approach and it can significantly outperform typical state-of-the-art denoising methods generally, especially for images corrupted by severe noise.