MATH MODELING

The new concept of a neural network for recognition and classification of pixel images

I. O. Grebenkin^a, N. A. Magnitskii^b, A. Yu. Chernyavskiy<sup>a

a</sup> Lomonosov Moscow State University, Faculty of Computational Mathematics and Cybernetics
^b Institute for Systems Analysis of Russian Academy of Sciences, Moscow

Abstract: In the work a new approach for the solving of a problem of recognition of black-and-white pixel images with use of artificial neural networks is offered. The algorithm of construction and training of artificial neural network LICS (Linear Characters Separation), concerning to multilayered perceptrons or perceptrons with the hidden layers is considered. The given algorithm has a number of advantages, among which - the minimal number of established parameters and, as consequence, an opportunity of work of algorithm in the whole class of tasks without preliminary 'adjustment' for each specific task, and also its presentation and simplicity.

Keywords: pixel image, neural network, perceptron, learning algorithm.

Bibliographic databases:

•