Adaptive Image Denoising in Mixed Gaussian-impulse Noise Using Weighted Nuclear Norm Minimization

Apart from additive white Gaussian noise, images are often troubled by impulse noise in image acquisition and transmission. In this paper, a two-stage adaptive image denoising algorithm is proposed for robust image recovery in a mixed Gaussian-impulse noise environment. The algorithm mainly based on improved Hampel identifier and weighted nuclear norm minimization (WNNM) method. The first stage is to adaptively identify whether the entries are corrupted by the impulse noise and then replace the outliers with some averages. In second stage, nonlocal self-similarity is investigated to explore low-rank properties of natural image and then reconstruct the original image based on the WNNM. Experimental results demonstrate that our algorithm provides a significant improvement over the classical Hampel identifier.