A high performance NEural MUlticlassifier System (NEMUS) is proposed, which is characterized by a great degree of modularity and filexibility, and is very efficient for demanding and generic pattern recognition applications. The NEMUS is composed of two stages. The first stage is comprised of several callifiers that operate in parallel, while the second stage is a Decision-Making Network (DM-Net) that performs the final classification task, combining the outputs of all the clasifiers of the first stage. In general, the inputs of each classifier are the features extracted from different Feature Extraction Methods and correspond to various levels of importance. The performance of the proposed NEUMUS is demosntrated by a shape recognition task of 2-D digitized objects, considering various levelsof shape distortions. Three different kind of features, which characterize a digitized object, are used: (a). Geometric features, (b). 1-D scaled normalized central moments and (c). The angles of a fast polygon approximation method.
展开▼