| 摘要 |
PROBLEM TO BE SOLVED: To recognize even a relatively small object character with high precision by smoothing the edge parts of a nonlinearly normalized character pattern and comparing the smoothed character pattern with character patterns in a recognition dictionary. SOLUTION: Time-series data of a character inputted on a tablet 1 are converted into bit map data, which are normalized. For the normalization, a redundant part of the input character is reduced to some extent by nonlinear normalization and then a character pattern of size as an object to be compared is obtained by linear normalization. After the normalization, directional features are extracted and the found directional feature pattern is defocused by a Gaussian filter. A character recognition part 3 matches the defocused patterns of directional features of characters generated by the defocusing process and previously generated in the character recognition dictionary 4 to find the similarities of all characters, and the character having the largest similarity is regarded as a recognition result.
|
