Research On Offline Handwritten Chinese Character Recognition

The reasearch on off-line handwritten Chinese character recognition is quite significant for the automatic Chinese character information processing and intelligent input of the new generation computer. In this thesis we engage on a further study on the application of the off-line handwritten Chinese character recognition into a special field, i.e. the handwritten recognition of the formal characters of Chinese numbers in the bank cheque.A well-equipped sample collection for written Chinese characters is one of the essential factors of commencing the research on handwritten Chinese character recognition. The establishment of Collection which contains the samples for the 19 formal characters of Chinese numbers, makes it possible to design the handwritten Chinese character recognition system for some special fields. This character collection consists of"零,壹,贰,叁,肆,伍,陆,柒,捌,玖,拾,佰,仟,万,亿 ,元,角,分,整', which is, to some extent, different from the Chinese character recognition in large amount in the traditional sense. All these provide feasibilities to the successful implement of this project.The major content of this thesis includes: the basic theory and methods of the off-line handwritten Chinese character recognition, the pre-work of the Chinese character images, the classification and recognition of handwritten Chinese character, the application of Neural Network(NN) into the off-line handwritten Chinese character recognition, the design for the usual plan concerning on the off-line handwritten Chinese character recognition system.As for the basic theory and methods of the character recognition, the thesis introduces the general methods and strategies in this field of optic character recognition(OCR) -statistical-decision algorithm based on mathematical characteristics of character and structure-decomposition algorithm based on characteristic structure of character. For the pre-work of the Chinese character images, the processes and methods of image binary, removing the noises, image thinning andimage standardization, which all are needed before the character recognition, are introduced. The classified algorithm of the character recognition includes the rude classification and particular classification, which both demand different strategies in the recognition correspondingly. For the application of NN in the Chinese character recognition, the Neural Network algorithm is also introduced. Meanwhile the BP Neural Network, required by the handwritten Chinese character recognition, is designed. By it the designed plan is emulated and testified.On the basis of the latest development in the field of Chinese character recognition, a Chinese character recognition system relying on the three-leveled recognition strategy is developed. On the first level, the characters are rudely classified by the traditional method of shadow-casting-coefficient transformation. On the second level, they are particularly classified according to the characteristics of the strokes density. On the third level, after integrating the most popular BP Neural Network algorithm, the recognition result is identified and output finally.