| 摘要 |
PROBLEM TO BE SOLVED: To realize a highly accurate character recognition result by using a statistical machine learning combining plural different character recognition methods.SOLUTION: Character recognition engines E1-EN respectively output plural evaluation values indicating respective degrees of applicability of an input character pattern being applicable to individual character codes; a feature vector generation part 14 generates a feature vector with the plural evaluation values as elements output from the character recognition engines E1-EN relative to the input character pattern; an identifier learning part 15 learns a category of the feature vector for each character code on the basis of the feature vector generated by the feature vector generation part 14 for the character pattern to which a character code is designated; and a character recognition part 18 determines a character code of an unknown character pattern on the basis of which category of the categories of feature information of each learned character code the feature vector of the unknown character pattern is applicable to. |
