CNN based on universal binary neurons: learning algorithm with error-correction and application to impulsive-noise filtering on gray-scale images
Naum N. Aizenberg, Igor N. Aizenberg, Georgy A. Krivosheev · 2002
In this paper we consider CNN based on universal binary neurons (UBN). Such an network is a very good base for solution of the different problems of image processing thanks to universal (absolutely) functionality of the UBN. New fast convergenced learning algorithm for UBN based on error-correction rule is presented. Solution of the XOR-problem on the single UBN is illustrated. Two functions for filtering of different kinds of impulsive noise (single impulses, combination of the single impulses and "scratches") are obtained. Templates for implementation of these functions on the single UBN are carried out by learning algorithm. Examples of impulsive noise filtering on the gray-scale images by the software simulator of the CNN are presented.