Noise variance estimation in digital images using iterative fuzzy procedure
Arianna Mencattini, Marcello Salmeri, S. Bertazzoni, A. Salsano · Cineca Institutional Research Information System (Tor Vergata University) · 2003
In the field of image processing, the noise estimation is an interesting issue in the filtering action. Almost all the filters, whose aim is to reduce the image degradation, need to know the noise content to optimize their performance. This paper presents a novel method, called IFP (Iterative Fuzzy Procedure), suitable to get a very good estimation of the noise variance, which represents the noise power, in the case the noise has a Gaussian distribution. The algorithm is an iterative procedure, tuned by a regression analysis, based on the fuzzy processing of some properties of the image measuring the matching between the theoretical luminance distribution and the local one. It has been tested on many images in a wide range of the noise variance. The achieved results, compared with those obtained with some other methods proposed in literature, show that the proposed approach is always accurate also in those situations in which the other ones often fail. Key-Words: Noise variance estimation, Fuzzy processing, Regression analysis.