Peer group switching filter for impulse noise reduction in color images

摘要

In this paper a novel approach to the problem of impulsive noise reduction in color images is presented. The proposed technique is based on the evaluation of the statistical properties of a sorted sequence of accumulated distances used for the calculation of the vector median. The detection of corrupted pixels is performed using the Fisher's linear discriminant working on the aggregated distances assigned to each of the pixels from the filtering window. The aim of the discriminant is to divide the pixels into two classes: a peer group of pixels similar to the vector median and a set of pixels consisting of outliers injected into the image by the noise process. The described filter enables reliable detection of impulses and its output switches between the vector median and the original, undisturbed pixel. In order to increase the filter's performance, a thresholding scheme is introduced to enhance the detail preserving abilities of the proposed noise attenuation scheme.