Fuzzy based self-similarity weight estimation in non-local means for gray-scale image de-noising

Abstract

Image de-noising in imaging systems is one of the most demanding issues that have been addressed in this work. Non-local mean filtering is gaining traction as a convincing method for image de-noising. Nevertheless, the performance of the traditional non-local mean filter is diminished by the computational expense of its weighted averaging process. In the proposed method, the similarity between the two patches of the image subspace is used to find the best weight for each pixel using a fuzzy inference algorithm. By using the similarity of non-local neighborhood pixels as an input antecedent in the input fuzzy set and the degree of weights as a consequent in the output fuzzy set, the fuzzy inference system which consists of IF-THEN rules, implication, and aggregation is employed. Eventually, de-fuzzification helps to measure how similar the pixels are to their non-local neighborhoods and estimates the fuzzy degrees of weight. Experiments on a variety of speckled noisy gray-scale images from standard and real public datasets demonstrate the improved performance of this fuzzy-based self-similar weight approximation in non-local mean filtering when compared to state-of-the-art image de-noising methods and sophisticated non-local mean methods in terms of standard evaluation metrics. Additionally, the suggested technique works satisfactorily when tested on actual speckled synthetic aperture radar images.