Fast and reliable noise level estimation based on local statistic

Noise level is an important premise of many image processing applications. This letter presents an automatic noise estimation method based on local statistic for additive white Gaussian noise (WGN). Analysis of the distribution of local variance shows that when local variances are not greater than the threshold that satisfies a special condition, their average is always linearly correlated with the real noise variance. Thus the actual noise variance can be obtained from these patches. Based on this idea, this letter provides an iterative process to select flat blocks, and estimates noise variance from these homogeneous patches using principal components analysis. Addressing challenges in noise estimation has major contributions to (1) studies on the distribution of local statistic and (2) an iterative process for choosing flat patches, which is the fundamental work of patch-based methods. The experiment results show that the proposed algorithm works well over a large range of visual content and noise conditions, and performs well in multiplicative noise. Compared with several conventional noise estimators, it yields best performance and faster running speed. (C) 2016 Elsevier B.V. All rights reserved.