Method and device to pattern recognition on statistical basis the invention relates to a method and device to pattern recognition on statistical basis, in which the relationship to each air-class of a class-set will be estimated with value

A method to pattern recognition on statistical basis, in which for a to-be-recognized object on basis of a full ensemble of two- or more-class classificator, the relationship to each aim-class of a class-set will be estimated with numerical value, the method gives a result through cascaded utilization of polynomial classificator, which is characterized in that: from all two- or more-class classificator through its estimation-vector-spectrum on a learning random sample, in which all to-be-recognized class-pattern are strongly represented enough, a selection of such two- or more-class classificators, whose estimations contribute with higher separation-relevance at the strongest to the minimization of a scalar measure that is computed through the estimation-vector-spectrum, is carried out; by means of the selected two- or more-class classificator, then the estimation-vectors will be built through an expanded learning random sample, from which the expanded feature-vectors will be generated through polynomial link