Research On Off-line Handwritten Chinese Character Recognition Based On Extreme Learning Machine

Chinese character recognition, as one of the technologies to automatic Chinese characters input, is an important interface for Chinese information processing. It can take an important part in many fields of our life, such as letter selecting, check recognition, report form disposing and hand-written manuscript auto-input. However, for Chinese characters are so much different when written by different people, hand-written Chinese character recognition is more difficult and becomes the main subject in the field of Optical Character Recognition.The main research content of this article includes:Chinese characters image preprocessing, Chinese characters feature vector extraction, and the important content is OV-ELM (Optimal Voting Extreme Learning Machine) which is the improved algorithm of ELM (Extreme Learning Machine) application to Chinese characters recognition.In this paper, Chinese characters image processing technologies are analyzed, including the sample image normalization, smoothing, character segmentation, image binaryzation, and refining processing. The analysis of the existing feature extraction method based on character recognition of Chinese characters, according to the characteristics of the small character set, this paper uses elastic meshing direction pixel probability distribution of feature extraction methods, extraction for ELM classifier input feature vector; in the classifier design, this paper designs the optimization algorithm of ELM based on the optimal voting mechanism OV-ELM, using OAO(One-against-one) or OAA(One-against-all) thought, put multiple classification problems for the degradation of a plurality of two classification problems to improve the classification accuracy rate, and the OV-ELM optimization of the traditional voting mechanism, the vote value to probabilistic manner, thereby avoiding the number of votes the same situation occurs. In this paper, the design of a series of experimental result can be seen, is presented based on elastic mesh feature extraction method for Chinese characters in the feature vector extraction results, using OV-ELM algorithm is compared with ELM algorithm to sacrifice a certain amount of training time, but still far superior to the traditional BP neural network, and in the classification accuracy and performance of more than ELM, SVM (Support Vector Machine).The overall classification framework in off-line handwritten Chinese characters recognition has achieved satisfactory classification performance.