Key Technology Of Clearing Of RMB Image Crown Word Number Recognition

RMB banknote is the legal currency of China and the main means for people to pay for something in daily life. Many disputes and criminal cases are related with RMB banknotes. As each RMB banknote has a unique crown word number, so the crown word number can be used as an identification of an RMB banknote.In recording the crown numbers of RMB banknotes and building a library of them, tracking and managing RMB banknotes can be effectively realized which has a great significance for maintaining the stability of the financial system and social harmony.In this paper, according to the technology of digital image processing and the knowledge of pattern recognition and combined with the features of the RMB banknotes themselves, auto recognition of RMB banknotes’crown word numbers is implemented. The main contents are as follows:(1) A pre-processing method combining the see-through image and the reflection image of a same RMB banknote is proposed here. Firstly, binarize the see-through image with an area-ration based method, then remove the OMRON rings in it with help of the corresponding reflection image which has total information of those OMRON rings. Then locate the crown word numbers region with prior knowledge and implement the tilt correction.(2) A character segmentation method combining the original see-through image with the image without OMRON rings is proposed. The image without OMRON rings has no interference around the characters and easy to be segmented, but some characters has lost information during pre-processing. As the original see-through image has total information of all the characters, and some may be adhered by the OMRON rings which would lead to hard segmentation, so consider to use the former image to implement the segmentation and record the borders of each character. Then segment the latter image with the borders and use its segmented characters as the resource of the latter recognition.(3) A method on8-direction edge feature combined with grid extraction is used here. Some common features of characters are introduced and analyzed at first. As grid can reduce the image dimensions while gradient can well describe an image, so combine these two features as the feature of the characters here.(4) The recognition of the crown word numbers based on template matching and SVM are implemented respectively. The experiments show that the algorithms in this paper reach a high recognition rate of the crown word numbers.