Font Classification for Multiple Font Character Recognition

摘要

The present invention has been invented to improve the amount of learning data increases and the character recognition rate decreases as the number of fonts to be recognized increases by effectively clustering fonts in a multi-font character recognizer.;In the implementation process of the present invention, first, a large number of fonts to be classified are first divided by a manufacturer, and then divided into a series such as Myungjo, Gothic, Palace, etc. To be selected. Secondly, the fonts set as criteria are expanded as much as possible while maintaining the shape of the letter using a neighbor relation search algorithm. Third, the average out ranged character ratio (ORCR) is calculated by using OR of fonts for each series and extended reference fonts. If the value exceeds the ORCR, the corresponding fonts are grouped into two groups. Reclassify as Fourthly, if the fonts classified by the above processes are ANDed for each group, they are classified into two or more groups by reclassifying only when the result type is not maintained to be distinguishable from other letters. By obtaining the training data, it is possible to drastically reduce the amount of training data than to obtain training data for each font.