Connected filters for noise removal

Abstract

A class of filters for edge-preserving noise removal is presented. The basic idea is that to retain connectedness in the filter, the new pixel value is calculated from the values in a local, connected area. The method described is related to the well-known KNN (k-nearest neighbors) method. The filter is shown to behave better than the KNN filter in most situations. The filter is conceptually very simple, and thus simple to use. Its computing efficiency is comparable to that of the KNN filter.<>