Author(s): Marek Kurzynski, Marcin Majak, Karol Puchala
Abstract: In the sequential recognition there exist dependencies among the successive objects to be classified. In this study two original multiclassifier (MC) systems for the sequential recognition are developed. In the first MC systems base classifiers are defined for particular steps of sequential recognition independently, whereas in the second MC system base classifiers classify an object at the current step on the base of its features and features of previous objects. Both MC systems in combining procedure uses original concept of meta-Bayes classifier and produces decision according to the Bayes rule. The performance of both MC systems were evaluated experimentally and compared with six state-of-the-art sequential recognition methods using computer generated data. Results obtained in experiments imply that MC system is effective approach, which improves recognition accuracy in sequential decision scheme.