PURPOSE: To improve a recognition accuracy by deciding a recognition result with the use of only a characteristic component reflecting the mutual difference among recognition candidate characters after extracting the recognition candidate character with the use of all characteristic components.

CONSTITUTION: The characteristic component of the characteristic quantity of a recognition object character and the character component of the standard characteristic quantity of each character stored in a dictionary 11 are collated in a main classification part 4 and plural recognition candidate characters for the recognition object character are obtained. At a key characteristic component decision part 5, a key characteristic component being a characteristic component suitable to indicate the mutual difference between the recognition candidate characters out of the character components is decided based on the dispersion of each of a standard characteristic in the recognition candidate character obtained at the main classification part 4. Next, the key characteristic component of the characteristic quantity of the recognition object character and the key characteristic component of the standard characteristic quantity of the recognition object character are collated at a detailed classification part 6 and the recognition result for the recognition object character is decided. Thus, a microscopic difference between similar characters can be detected and the recognition accuracy of the character can be improved.