Research On Recognition Method Of Paper Currency Number

The paper currency recognition equipment on the market such as cash registers, ATMmachines and sorter recognizes the paper currency by magnetic characteristics, opticalcharacteristics, etc, but they can not recognize the high imitation counterfeit. Every papercurrency has a sequence number, and the number is unique, so it can be used to distinguishthe paper currency. If the equipments can be added number recognition unit, recognize andrecord the number automatically when tying the paper currency, use unified managementon the paper currency, it will has wide practical value and broad application prospects.For the high requirements of the processing speed and recognition rate of the papercurrency number recognition in practical engineering applications, this paper studies therapid recognition algorithm of paper currency number, and applies it to the actual papercurrency recognition equipment.Firstly, this paper studies the image preprocessing algorithm, including zoom andbinarization. For the requirements of fast and accurate, this paper uses double linearinterpolation to image zoom, and uses iterative method to binarization. Secondly, thispaper studies the algorithm of character segmentation, using the three-step projectionmethod to once segmentation, and twice segmentation is used to solve the segmentationproblem caused by the non-uniform gray distributions image to achieve precise positioningof each character. Lastly, this paper studies number recognition method, this paper uses themethod based on structural features of the characters. Meanwhile, according to thecharacteristics of the RMB number, this paper has resolved the alphanumeric transpositioneffectively.The algorithm used in this paper does not require image denoising and tilt correction,and it also avoids the normalization process, so it saves much time and achieves goodrecognition effect.This paper uses the algorithm to test actual collected images under PC, the algorithmhas been embedded into the sorter so that the sorter can achieve real-time acquisition andreal-time recognition,10,000actual collected images have been tested in this paper, the recognition rate is96.3%,and the recognition time is10ms. The algorithm can meet thereal-time requirements and has a high recognition rate, so it can meet the needs ofindustrial production.