Recognition And Application Of Handwritten Tibetan Character

Tibetan is one of the major ethnic minorities in China,and the population of Tibetan language is about 8 million.Tibetan literature is so vast that it is rich in collections.Tibetan literature is also China's second largest language and cultural heritage with a long history and a variety of forms.However,due to the late information construction in Tibetan areas,there are relatively few researches on Tibetan recognition,especially the recognition of handwritten Tibetan.There is no mature Tibetan handwritten input software on the market,and the communication on the Internet is also dependent on the input of voice and keyboard.Therefore,conquering the difficulties of Tibetan recognition and promoting the development of Tibetan recognition technology are not only of great significance to the inheritance of ethnic cultures in China,but also closely related to the daily life of Tibetan compatriots.This thesis takes the Tibetan character as the research object,and carries out the theoretical and experimental research on the key technologies such as preprocessing,feature extraction and classifier design.Besides,this thesis realizes the handwritten Tibetan input system applicable to mobile platform.The specific work is as follows:1.The background and current research status of handwritten Tibetan recognition technology are introduced in detail,and the feasibility of handwritten Tibetan recognition based on Tibetan alphabets,syllables and characters is analyzed respectively.This thesis uses Tibetan character as the research object.According to the national standard of Tibetan characters,which are published in China,we count 663 local Tibetan characters.By developing handwritten Tibetan character collection software,this thesis collect 60 sets of handwritten Tibetan character samples,and establish a database of 39780 handwritten Tibetan character samples.2.In the preprocessing step,tilt correction based on the Hough transform and nonlinear normalization based on the point density are used,combined with the interpolation and resampling technique,to preserve the original information of the handwritten Tibetan character,and filter out redundant information,ensuring the effectiveness of feature extraction.In terms of feature extraction,the directional elements features and the Gabor features of handwritten Tibetan characters are studied respectively.This thesis proposes a feature extraction algorithm that fuses online and offline features,to make use of the good complementarity between online features and offline features.Through the full use of online and offline information of Tibetan character,the algorithm provides a more reliable classification basis for classifiers and improves the overall recognition rate of handwritten Tibetan characters.3.In the aspect of classifier design,we use the two-level classifier cascade algorithm,which effectively utilizes the fast classification speed of Euclidean distance classifier and the good subclassification ability of MQDF classifier.In view of the high similarity of handwritten Tibetan characters,this thesis puts forward a candidate rearrangement algorithm that separates upper vowels and converts local information into global information,combining with the characteristics of handwritten Tibetan characters.This algorithm amplifies the similar parts of characters to improve the system's recognition performance.4.This thesis transplants the recognition algorithm with good performance from PC terminal to the mobile terminal,and implements a handwritten Tibetan character input system which can be applied to mobile platform.By adopting the framework of IMF,we have developed a software of handwritten Tibetan input method based on Android,which can satisfy the daily use of the Tibetan compatriots.