The Method Of Serial Number Identifying On RMB Banknote

OCR(Optical Character Recognition) is an important branch of pattern recognition. Many theories and methods have been studied and proposed in OCR area to this end. However , the applications and popularizations of OCR techniques cannot be satisfied in long time. The reason of it is that there are different requirements for system in different lines, so the developments and spreads of OCR industry must be lines oriented. In this paper, a character recognition system in OCR, which is financial oriented has been studied.Document segmentation of this system gets the positions of every word by horizontal projection and vertical projection. The words are segmented into characters or conglutination characters by connective areas searching, which is regarded as the elementary recognition unit. The task of this article is document segmentation, character feature extraction, classifier designing and first step post process. Document segmentation is regarded as one of the preprocess in this article to be explained.Before recognition, the initial image should be binarization, removing blot and white edge and normalization. Thus all images can be at the same standard with less interferential factors. This can build foundation for latter feature extraction and recognition.In the article, several features which are used widely in OCR, such as DEF feature, girding feature, through feature and distance classifiers are explained and compared, and many experiments are did. Based on the characteristics of the English characters, through feature and grading feature are selected and blot distance is used as the classifier of this system. In order to enhance the classification ability, add different coefficients at character vector, and the coefficients are decided by experiments.Last the article make the network with BP Algorithm to recognize the character, the strongpoint is great self-studying ability and fast speed.