Recognition Of Handwritten Tibetan Syllable Words

Nowadays,with the rapid development of science and technology and information technology,the research of handwritten Tibetan recognition is still in its infancy,Tibetan compatriots' mobile phone input method software is still in full keyboard mode,Tibetan literature digitization work still needs multitude.From the Tibetan compatriots living needs,the ethnic cultural heritage,all cannot do without the development of Tibetan language handwriting recognition technology.Therefore,it is of great significance to protect the intangible cultural heritage of ethnic minorities by research the handwriting recognition technology of Tibetan language,which is the bounden duty of Chinese character recognition workers.This paper introduces the current research situation of Tibetan recognition,and deeply analyzes the features and difficulties of Tibetan.Due to the practical application of Tibetan syllable word as writing unit,this paper first makes a systematic theoretical and experimental study on the recognition of the handwriting Tibetan syllable words.In this paper,the method of word recognition based on feature extraction and deep learning are applied to the study of Tibetan syllable word recognition,and the main results are as follows:1.With the help of the Tibetan compatriots,this paper makes a statistical analysis of 579 commonly used Tibetan syllable words according to the data of the dynamic ratio of the letter combinations of Tibetan syllables,collects 60 sets of handwritten Tibetan syllable words,and establishes a database containing 34740 handwritten Tibetan syllable words.2.This paper uses feature extraction to recognize handwritten Tibetan syllable words.In the pre-processing step,in order to preserve the structural information of Tibetan syllable words,this paper presents a linear normalization method,which normalizes the upper and lower parts of the Tibetan syllable.This paper combines this method with nonlinear normalization,smoothing,interpolation and resampling,it retains the maximum original information of the handwritten Tibetan syllable word,filters out redundant information,which is convenient for feature extraction and recognition.In the feature extraction step,according to the characteristics of handwritten Tibetan syllable words,this paper proposes a method that combines the features of upper vowels and the eight directional elements as the features of handwriting recognition.During the classifier design procedure of feature extraction method for character recognition,in order to solve the problem of similar words and low recognition rate due to the small difference between the vowels of the Tibetan syllable words,and consider the special development environment of the mobile platform,this paper presents a design scheme of the three level classifier,which takes the upper vowel feature as a rough classifier,and the Euclidean distance and the MQDF as a precise classifier.3.This paper compares the recognition effect of Euclidean distance and three level classifier.When using the three level classifier,the average recognition rate of the first three is about 94.18%,and the average recognition rate of the first five is about 96.45%.Experiments show that this algorithm has better recognition effect.This paper tests the performance of the algorithm based on feature extraction,and applies it in the mobile terminal,it uses Google's Android Input Method Framework for the development,and implements the software of full handwritten Tibetan input software based on Android.4.This paper uses the method of deep learning in handwritten Tibetan syllable word recognition.This paper improves on the network structure of Le Net-5 model,which is a classic handwritten numeral recognition network,by adjusting the structure and parameters of the network,it achieves a optimal performance,the experimental results show that the highest recognition rate reached 91.2%.