Research Of The Technology Of Online Handwriting Mongolia Words Recognition

Online handwring recognition is fundamentally a pattern classification task, the objective is to take an input pattern, and the handwritten signals collected online via a digitizing device, and classify it as one of a pre-specified set of words (i.e., the system's lexicon). Because of large variation of handwriting, exact recognition is very difficult. Especially the connectivity between the characters, make the recognition more difficult.During recent years, the task of online handwriting recognition has gained an immense importance in every day applications, mainly due to the increasing popularity of the personal digital assistant (PDA). Now there are many products of online handwriting recognition of Chinese characres and West characters. Mongolia language is very popular among the Mongolia people in the North China, so the research of online handwriting Mongolia words recognition has a far-reaching meaning about developing the Minority information technology and national culture. Mongolia word is a kind of spelling characters, which has a very special written structure different from Chinese and English characters. It is written from left to right, from top to bottom, all letters are connected together to form a vertical backbone, and every letter may have different shapes in different positions. All these characteristics bring many difficulties to recognition.This paper primarily discussed Online Handwriting Recognition methods for Mongolia words. We introduced the multiple classifiers which were built on different feature sets. We make use of Online and Offline information for feature selection and character segmentation. And online feature applied to HMM Classifier, offline feature applied to Nearest Neighbor Classifier. Our classification combined both HMM model and Nearest Neighbor Classifier based on DTW distance (Dynamic Time Warping), so we called Multiple Classifier. Experimental results show that writer-dependent words achieve recognition rates above 95%. And unconstrained words achieve recognition rates from 83%~92%. Our system run well, and can apply to application to some extent.