Recognition texture patterns in digital images using an hybrid artificial network

Abstract

This work presents a mechanism of images indexation based on textural features applying artificial neural network. Attributes originated from Haralick descriptors based on co-occurrence matrix, are applied in order to classify the textures. An hybrid implementation of neural network is used in the recognition of different texture patterns, which firstly propitiates a initial gathering of them by using an unsupervised model (self-organization map [SOM]) and secondly by using a supervised model (learning vector quantization [LVQ]) aiming to enhance the patterns classes segmentation previously gathered from the application of the unsupervised model.