/Interperiodica ” (Russia). Selection of Typical Objects in Classes for Recognition Problems

—Discrete recognition procedures based on a search for sets of feature values that are not encountered in the feature descriptions of training objects are considered. The constructed recognition procedures are com-pared with classical procedures for real-life applied problems. An approach to improving the performance of recognition algorithms based on selecting training objects typical for each class is examined. A fast method for calculating estimates in voting over representative sets for the cross-validation procedure is suggested