HMM-Based Recognition Research On Online Handwriting Uyghur

With the development of the Xinjiang information technology, the demand for machine recognition of handwriting Uyghur is increasingly urgent. It not only is important on the use of ancient Uyghur literature, but also essential for today’s economic development. There have been some researchers studied in it. But there has not been a relatively mature research results. In the final analysis is the writing characteristic of Uyghur text to identify the difficulties, and other effects to the recognition results. Uyghur is similar to the Arabic, so we can use the technology on the Arabic recognition to the Uyghur. At the same time, the methods for the recognition of Uyghur can also be used in the research of Arabic. This experiment is complicated the on-line handwriting recognition of Uyghur.In this paper, on the bases of the characteristics of the Uygur, this experiment has done the analysis, research and experiments of the character recognition technology in the recognition part. By building and classing the HMMs of the characters, building the dictionaries of the samples and building the network for recognition, we have worked out the recognition by using HTK tools. In the training phase, the sample words were cut into the letters manually, and then extracting the features of the samples, constructing HMM models for the primitives of letters, and finally embedded the HMM models into the recognition dictionaries in the recognition network. The dictionaries for the recognition include three parts:dictionary with delayed strokes, dictionary without delayed strokes and a spare dictionary. In the recognition stage, the word parts will be the primitives for recognition, through eliminating their delayed strokes and then using the recognition dictionaries and network to be recognized. Result shows the recognition accuracy rate is up to 90%.