Segmentação de imagens por classificação de cores: uma abordagem neural. / Image segmentation by color classification: a neural approach.

AUTOR(ES)
DATA DE PUBLICAÇÃO

2000

RESUMO

The present work approaches the segmentation of colored images through the process of color classification, i.e., the segmentation of images based on the color attribute of pixels. We look for a color classification as close as possible of human classification. In other words, we look for a robust classification with respect to the variation of illumination and color brightness, which tries to be tolerant to errors in the sampling process. We may find such kind of problems is various practical situations, for instance, situations that is influenced from the environment in the application domain: the robotic soccer. With regard to this problem, there are still diverse questions that remain unsolved, such as color representation form and type of classifier which maximizes the classification performance. In fact, classic models have shown to be inadequate in this context, in general, stimulating us to investigate new solutions. In our work, we present a classifier using one technique that has shown great applicability in this scope: the artificial neural networks. In order to obtain a correct generalization in the network, we faced the necessity to build a methodology to supply examples in the training phase of the network. In short, we model and implement a classifier, while searching to asses about its generalization power in different contexts and in the universe of training, so as to determine the set of factors (system of representation of colors, methodology of supply of examples and architecture of network) that maximizes its performance.

ASSUNTO(S)

visão computacional pattern recognition artificial intelligence image processing redes neurais artificiais colors reconhecimento de padrões artificial neural networks processamento de imagens inteligência artificial computer vision cores

Documentos Relacionados