| 摘要 |
PURPOSE:To improve precision of discrimination between a possible category by selecting a feature on the basis of an F ratio (inter-class dispersion/in-class dispersion) using plural dictionaries. CONSTITUTION:The feature of an input character is extracted and inputted to the 1st deciding circuit 5 and the distance between possible category read out of the plural dictionaries 4 is calculated to select two possible categories 1 and 2. In this case, ''KYU'' as the possible category 1 and ''KAN'' as the possible category 2 (''KYU'' and ''KAN'' are similarly shaped Chinese Chracter.) are selected and the dictionaries 6 and 7 are used for selecting the possible categories 1 and 2. Then, the F ratio (between inter-category variance and in-category variance for every feature) between the dictionaries 6 and 7 used as mentioned above is calculated; and such a feature that the F ratio is maximum is used to calculate the distance between the input character and plural possible categories, and the category having the least distance is outputted as a final result. Thus, deformation associated with the handwriting is dealt with to improve the discrimination precision. |
