Adaptive Iterative Low-Rank for Real Image Denoising

Noise distribution of real images is often unknown, and generally has no regularity, therefore the denoising of real images has always been challenging. Based on the prior knowledge that there are a large number of self-similar patches within real images, this paper considers that the similarity and the number of similar patches in real images are different, and then proposes a similarity function to estimate texture features of patch groups. After clustering similar patches by using Gaussian mixture model (GMM), an efficient adaptive iterative scheme for low-rank denoising is proposed to enhance sparsity of patch group and prevent over-smoothness simultaneously.