The Research On Off-line Handwritten Chinese Character Recognition Based On Thinning

Chinese characters as the unique form of the pictograph have been considered as the most ancient characters of the world, which has more than 6000 years old. As the basic communication tool, Chinese characters possess such qualities as being abundant in quantity, complex in structure, and big amount in similar spellings; therefore, handwritten Chinese character recognition is regarded as the most development form of model recognition, in which off-line handwritten Chinese character recognition is currently referring to the most difficulty and challenging research topic. Additionally, concerning Office Automation and Machine Translation, off-line handwritten Chinese character recognition has a huge potential in application, which has caught attentions over the world. To draw a conclusion, the research on off-line handwritten Chinese character recognition not only is of great importance in theoretical value, but also contains significant value in use.Off-line handwritten Chinese character recognition has generally been divided into: pre-operation, feature extraction, classification methods. This thesis concerning the above procedures mainly involves the following aspects:(1) In pre-operation step, make use of the neighbour average filter realize the smoothing and rough removing process, and in accord with the different paper sources, various methods are taken to deal with the image binarization. And then, the author studies on the tilt correction process, line and character segmentation, and standardization, and cope with the corresponding facts with specific techniques, which serves as the firm foundation of the followed feature extraction process.(2) The author also reviews the commonly used thinning methods and with the particular quality of the system in this thesis, provides the improvement based on SPTA thinning methods, which is proved to be effective(3) According to the classification of feature extraction, the author fully introduces the full-scale statistical characteristics, partial statistical characteristics, advantages and disadvantages of structure, and also adopting the extraction method for probability distribution of flexible meshing pixel. (4) The design of multi classification methods is introduced as well, and the detailed description on classification and recognition, learning and training of the BP neutral network is also available in this thesis.This paper presents the deep research, which based on off-line handwritten Chinese character recognition theory, and self-developed operation system of student achievement. This topic belongs to small Character set of the handwritten Chinese character recognition, which is much distinguished from traditional big set system serving the possibility of the success of this thesis. In the system, the author builds a sample-base which contains "excellent, good, average, bad" handwritten Chinese characters. The system recognizes the paper-making record cards with the help of electronic scanning, which realizes the automated management on student achievement and improves the efficiency of office performance.Throughout the testing on the experimental samples, the system has a 96.25% correct recognition rate on the samples. Although this is just an experimental model, with much potential space, the attempt is worthy enough for future research on the successful application of off-line handwritten Chinese character recognition.