METHOD FOR PRODUCING DECISION TREE IN PATTERN RECOGNITION

摘要

PURPOSE:To produce a dedcision tree used in pattern recognition with high speed by taking into account fluctuations in the calculation time of feature amounts and calculating these amounts computed in parallel in the preparation of the decision tree. CONSTITUTION:A classification unit 2 classifies the category within the class for each feature amount according to the relative magnitude of these amounts. An occurrence frequency memory part 4 stores the occurrence frequency of discrimination object corresponding to the respective categories. A memory part 5 for storing the mean time of the feature amount stores the mean time of other calculation time for feature amount excluding the calculation time for the feature amount when assumed that a given feature amount has already been calculated. A calculation part 6 for the evaluation of feature amount 6 calculates for each such amount the maximum recognition time and comprehen sive recognition time and stores them in a memory part 7 for evaluation of feature amouts. By this method, the maximum recognition time and comprehen sive recognition time can be reduced.