Evaluation of the use of two different models of neural networks in the classification of images based on textural features

Abstract

This work explores the classification of textures by artificial neural networks. Two different models of neural networks are used: one supervised (multiple layers perceptron) and the other non-supervised (Kohonen auto-organizable maps). In both cases, the attributes that describe the textures and those ones used to classify them, result from statistic approaches of first and second orders. In this study, a comparative analysis between the experimental results of each model of networks is carried out.