| 摘要 |
<p>PURPOSE:To improve a character recognizing performance by calculating the composite similarity of a feature vector and a dictionary vector to obtain the similarity and deciding a character recognition result from the similarity. CONSTITUTION:A feature extracting part 13 consists of a 2X2 vicinity pattern extracting part 21, a vector calculating part 22 and a normalizing part 23. A vicinity pattern consisting of 2X2 dots is extracted from an input pattern to calculate the feature vector based on frequency information at every position of a specific pattern of the vicinity patterns. The feature vector extracted in the extracting part 13 is applied to a composite similarity calculating part 14. In the calculating part 14, the composite similarity of a standard vector stored in a recognition dictionary 15 and the extracted feature vector is operated to obtain the composite similarity on respective categories. Accordingly, a high character recognition ratio can be obtained.</p> |
