| 摘要 |
<p>PURPOSE:To perform correct character recognition, by extracting a subpattern representing a stroke in each desired direction from a character pattern and the subpattern other than the subpatterns mentioned above, and performing recognitions after preparing a feature matrix from the extracted subpatterns. CONSTITUTION:An optical singal input 1 is converted into a digital electric signal through a photoelectric conversion 2 and planted into a pattern register 3, and, at the same time, a stroke width calculation 4 is made. A vertical pattern extracting section 5 performs vertical scanning throughout the pattern register 3, and extracts vertical subpatterns from the relation between the continuous length of the black bits and the stroke width calculated at the calculating section 4. In the same way, the extracting section 5 extracts subpatterns in the horizontal, obliquely right, and obliquely left direction. Moreover, the section 5 also extracts underfined subpatterns with black bits which do not belong to all of the subpatterns mentioned above. A character frame detecting section 10 detects the outer frame of the character patterns and extracts 12 characteristic matrixes together with the output of all subpatterns through a character frame separation deciding section 11. The feature matrix thus extracted is compared with a standard character mask at a recognizing section 13, and outputted 14.</p> |
