PURPOSE: To satisfactorily recognize an inputted blur and faint character pattern in terms of detecting the shapes of a standard character pattern and an inputted character pattern by adding the deformation of a prescribed shape.

CONSTITUTION: Estimation steps 5 and 7 in the flow of a broad classification method in terms of character recognition deform shapes. KANJI (Chinese character) 'KO', for instance, is printed blurredly because the widths of S1 and S2 are narrow, and there is concern that the shape between them may not be detected. If the widths of the S1 and S2 are smaller than a distance δ, a step 5 in a dictionary generating mode deforms the character to neglect said widths, and adds codes formed by that deformation to classifications in the dictionary. KANJI 'KUNI', for instance, is prone to develop gaps such as S3 and S4 due to fait printing. When the widths of the S3 and S4 are smaller than the distance δ, at the step 7 in a character identifying mode, deforms the character to neglect said widths, the characters classified into codes in the dictionary whose codes coincide with the codes formed by such deformation, are added to the selection of main classification.