| 摘要 |
PURPOSE:To more accurately recognize a character even in case of a document in which different types of one same character are used by registering plural kinds of standard feature amounts of each character in a dictionary for recognition. CONSTITUTION:A character segmenting part 2 segments a recognition object character image in square shape from an image inputted through an image input part 1, and a feature amount calculation part 3 obtains the directivity, number, and position, etc., of the stroke of a recognition-object character image obtained by the part 2, as the feature amounts of said image. Also, a recognition candidate extraction part 4 collates the feature amounts of the recognition object character image obtained by the part 3 with the standard characteristic amounts of each character registered in the dictionary 6 for recognition and extracts recognition candidate characters in the order of degree of resemblance. A learning part 5, based on the distances between the feature amounts of thus obtained recognition object character images and the standard feature amounts of learning object characters which are characters corresponding to the recognition character images stored in the dictionary 6 for recognition, updates the standard feature amount of a currently attended learning object character, or newly stores it. A display part 7 displays a recognition result obtained by the extraction part 4.
|
