TÉCNICAS DE VISÃO COMPUTACIONAL A PARTIR DA DETECÇÃO DE PADRÕES COMPRIMIDOS EM IMAGENS DIGITAIS

Abstract

Computer vision, the science that seeks to emulate aspects of human vision through the use of hardware and/or software, had massive contribution in past decades. This works presents a new data compression based and parameter-free approach for the construction of a generic model of representation of images through the adaptation and extension of the DAMICORE methodology. An application was developed, and tests ran with some image sets, good results were obtained, and in general, good generalization capacity is shown. Though good results were not obtained in all experiment categories, possible modifications in the technique are presented as future work, besides ideas about the extension of the technique to classification tasks and to other kinds of data.