PURPOSE: To improve the accuracy of recognition by executing the recognition processing of the same input character picture unitedly on the basis of plural feature data extracted at different levels.
CONSTITUTION: A feature extracting part 4 is controlled by a feature extraction parameter switching part 11 and executes the feature extraction processing of one input character picture on the basis of plural feature extraction parameters P1WP3. The extracted input feature data are stored in an input feature data storing part 5 respectively. A distance calculating part 6 matches a pair of input feature data X1WX3 with reference feature data stored in a reference feature data storage part 7 to calculate the distance and selects every three categories. A weighting processing part 12 sets different weight values to three categories selected by the calculation part 6 in respective data X1WX3. A weight appreciation part 13 arranges the categories in the descending order of weight. A recognition output part 8 selects the category having the maximum weight and outputs the category as a recognized result.