Enhancing efficiency and speed of an off-line classifier employed for on-line handwriting recognition of a large character set

This paper proposes a new approach to acceleratingspeed and increasing the recognition rate of an off-linerecognizer employed for on-line handwriting recognitionof Japanese characters. All training patterns are dividedaccording their stroke number to several groups and onesingle recognizer is dedicated for each group of patterns.Since a number of categories for a single recognizer issmaller, the speed and accuracy improves. First, we makethe model of a recognizer and show that our method cantheoretically accelerate its recognition speed to 45% of theoriginal time. Then, we employ the method to a practicallyused off-line recognizer with the result that the recognitionrate is increased from 90.73% to 91.60% and therecognition time is reduced to only 49.73% of the originalone. Another benefit of our new approach is highscalability so that the recognizer can be optimized forspeed and size or for the best accuracy.