Design And Implementation Of Online Handwritten NÇšshu Characters Recognition System

NÇšshu is the unique surviving characters and an ancient literature in the world.Therefore, we must take advantage of the information technology to protect it. As areal-time recognition method, online handwritten recognition has high efficiency ofrecognition and characters could be recognized when naturally written on ahandwritten board, which plays an important role in protecting this precious culturalheritage. In addition, there are so many kinds of words in NÇšshu and appears to bevery artistic and peculiar that gives rise to be recognized difficultly. The research onthe online handwritten NÇšshu character recognition has a very important theoreticaland social role.According to the problem of online handwritten NÇšshu character recognition,some works have been carried out after anglicizing the characteristics of onlinehandwritten NÇšshu characters. We design a root-radical-stroke-based onlinehandwritten character recognition classifier and implement a stable online handwrittencharacter recognition system which has been put into use effetely.The main research work and features in this thesis are as follows:1) In the light of the characteristics of online handwritten NÇšshu characters, wepropose a structure-based recognize method and design a root-radical-stroke-basedonline handwritten character recognition classifier;2) Linear normalization method has been used for preprocessing. As a result of somany curved strokes in NÇšshu character, we propose an improved secondary searchmethod and combine with the line approximation method to extract feature points forstroke recognition by dynamic programming into stroke feature dictionary, then applythe connected domain into radical segmentation, rearrange the stroke order for radicalrecognizing, and utilize the long stroke feature and the strokes clustering informationto classify the Structure of roots in NÇšshu character. Finally, we make use of theradical and root mapping into recognition; other in-depth technology has beenresearched in this paper, such as distinction of similar characters, broken andconnected strokes.3) The experimental results show that the first-time recognition rate in ourapproach has been increased by16.4%than the ordinary online handwritten characterrecognition method based on strokes and stroke order; the ten-time recognition ratecould be reached as high as96.5%.