Blind estimation of noise level based on pixels values prediction

Abstract

Noise parameters estimation is required in various stages of digital image processing. Many efficient algorithms of noise estimation were proposed during last two decades. However, most of these algorithms are efficient only for a specific type of noise for which they are designed. For example, methods of variance estimation of additive white Gaussian noise (AWGN) will not work in the case of additive colored Gaussian noise (ACGN) or, in general, in the case of a noise with non AWGN distribution. In this paper, a totally blind method of noise level estimation is proposed. For a given image, a distorted image with a discarded portion of pixels (around 10%) is generated. Then an inpainting (or impulse noise removal) method is applied to recover those discarded pixels values. The difference between the true and recovered pixel values is used to robustly estimate image noise level. The algorithm is applied for different image scales to estimate a noise spectrum. In this paper, we propose a convolutional neural network called PIXPNet for effective prediction of values of missing pixels. A comparative analysis confirms that the proposed PIXPNet provides smallest error of recovered pixel values among all existing methods. A good efficiency of application of the proposed method in both AWGN and spatially correlated noise suppression is demonstrated.