MORAES, R. M.; http://lattes.cnpq.br/7925449690046513; MORAES, Ronei Marcos de.
Resumo:
Systems for contextual classification of images are known in the specialized literature for their good accuracy of results and slowness of computation, due to the fact that a large number of computations is needed at each interaction. For this reason, very few attempts have been made to produce practical or comercial implementations for this class of algorithms. In Brazil, the only national image processing systems comercialy avaiable does not offer a method of contextual image classification to their users. With the intention to fill this lack of a high precision classification method at the SITIM system - developted by INPE and comercialized by ENGESPAÇO of São José dos Campos - São Paulo and,
to verify the feasibility of a practical implementation of contextual algorithms, a bayesian classification method was studied, improved and implemented in this work. This implementation results are presented and the system is tested in comparison with a statistical method of classification by maximum likelihood, already implemented at the SITIM system.