New Neural Networks for the Affinity Functions of Binary Images with Binary and Bipolar Components Determining

Abstract

The Hamming neural network is an effective tool for solving problems of recognition and classification of objects, the components of which are encoded using a binary bipolar alphabet, and as a measure of the objects’ proximity the difference between the number of identical bipolar components which compared include objects and the Hamming distance between them are used. However, the Hamming neural network cannot be used to solve these problems if the input network object (image or vector) is at the same minimum distance from two or more reference objects, which are stored in the weights of the connections of the Hamming network neurons, and if the components of the compared vectors are encoded using a binary alphabet. It also cannot be used to assess the affinity (proximity) binary vectors using the functions of Jaccard, Sokal and Michener, Kulchitsky, etc. These source network Hamming disadvantages are overcome by improving the architecture and its operation algorithms. One of the disadvantages of discrete neural networks is that binary neural networks perceive the income data only when it’s coded in binary or bipolar way. Thereby there is a specific apartness between computer systems based on the neural networks with different information coding. Therefore, developed neural network that is equally effective for any function of two kinds of coding information. This allows to eliminate the indicated disadvantage of the Hamming neural network and expand the scope of discrete neural networks application for solving problems of recognition and classification using proximity functions for discrete objects with binary coding of their components.