Shape quantification of handprinted numerals using recognition results and its application

This paper considers the verification of the handwriting characteristics of characters and its application to the error detection through the quantification of handprinted numerals. It is shown first that the handprinted numerals can be quantified by applying the third model of the quantification theory for the frequency information of the candidate subcategories obtained from the recognition system. Then it is verified quantitatively as the writing characteristics using the result of quantification that the same writer writes the characters belonging to the same category in the same character shape, and there exists a correlation between different categories. Applying multiple regression analysis, character quantity of one category is predicted from character quantity of other categories written by the same writer.