Design And Realiazation Of CF1000 Paper Currency Sorter Image Recognition System

The paper currency sorter is a mechanical and electrical integration of the financial control equipment, which can make the counting banknotes, counting, identifying broken, old and new and different geometry, different versions of the banknotes at the same time. The paper currency sorter can improve the speed of the efficiency of cash handling for financial industry, make the work of banknotes classification too quickly and effectively. CF1000 currency sorter of this research project which has two note ports is a small paper currency sorter .It has small, fast running banknotes and other characteristics, mainly for the RMB currency, it is suitable for the promotion and using of domestic financial field.This paper mainly designs the system of image recognition for CF1000 paper currency sorter. The system of image consistis of notes image preprocessing, face and vaule of banknotes recognition, version of banknotes identification, testing new and old banknote, and defect detection. In the image preprocessing module, according to the image quality of the image sensor acquisition,compensating for the brightness of the image sensor acquisition to improve the quality of notes image and using the method of compensating the image and geting the image at the same time. At the same time using a rapid method of inspection side to locate thelocation of banknoke in the image.In banknotes recognition module, it designs the method of feature extraction and classifier of recognition face and value of banknotes. According to quality of image, using the method of template matching to identificate the face and value of banknote. The method of template matching for feature extraction is based on Feature grid of the local characteristics of the image. At the same, according to the width features of banknotes, it designs nominal classification. In the final, the method of identification used the combination of template matching and nominal classification,it improves the accuracy of the system.In the module of processing banknotes, for the version of the discriminant, the method based on gradient of feature area. According to characteristics of distinguish between versions of the regional, it uses the method which based on gradient of feature area to classify versions of the banknotes. The design of the defect detection uses banknotes on the characteristics of visible light through the low. In the detection of the old and new banknotes, according to the performance of the machine itself,it designs the method which combination of based on histogram of feature area in banknote a blank area of banknotes. It Improves the stability of old and new identification.Image recognition system that this paper designed has been applied to the CF1000 currency sorter, and achieved good results. The recognition accuracy rate of value and face has reached 99.8%, the rate of Version identification of accuracy has reached 100%, the error of old and new detection has reached 10 or less, the error of defect detection has reached 1 or less.