Automatic allograph categorization based on stroke clustering for online handwritten Japanese character recognition

For the construction of a recognition dictionary that includes various writing styles, an automatic method for categorizing writing styles of characters (allographs) is proposed. In the first step of allograph categorization, handwritten strokes contained in training data are categorized to obtain prototype strokes. These strokes are used to categorize handwritten characters and thus obtain allographs. In this approach, allographs share common prototype strokes. This makes it possible to reduce the dictionary size and computation time needed for recognition. Allograph dictionaries for 2321 categories were experimentally constructed by using handwritten characters produced by 121 writers. Recognition experiments using these dictionaries were carried out to determine the relationship between the number of allographs and the recognition accuracy