Нейромережеві методи структурної класифікації зображень

Abstract

The given work is devoted to solving the problem of invariant recognition of visual objects by using methods that classify descriptors of singular points of the image via Kohonen network. A classification method using a variable number of adapted neurons is proposed. The criteria for assessing the level of correct classification are analyzed. Software modeling of the proposed method is carried out. The results of the software modeling of the network using different variants of processing and formation of classes and cluster centers are presented.