PROBLEM TO BE SOLVED: To provide a technique capable of easily constructing a classifier for performing image classification with high accuracy.;SOLUTION: A frequency distribution data generation part 332 generates frequency distribution data 82 showing a frequency in each section obtained by discretizing values of a feature amount axis for each class in each feature amount axis on the basis of plural teacher data 81. A classifier generation part 333 generates a classifier 330 for classifying an image on the basis of the frequency distribution data 82. A frequency distribution data correction part 334 specifies a correspondence section corresponding to the value of a feature amount belonging to the teacher data 81 in each feature amount axis about the teacher data 81 that is erroneously classified when classified by the classifier 330 in the plural teacher data 81, and corrects the frequency distribution data 82 so as to increase the frequency of a class taught by the teacher data 81 in the specified correspondence section. A repetition control part 335 causes the frequency distribution data correction part 334 to correct the frequency distribution data 82 once or more times.;SELECTED DRAWING: Figure 3;COPYRIGHT: (C)2019,JPO&INPIT
展开▼