Research Of RMB Optical Character Recognition Algorithm

With the high-tech of producing counterfeit money, it is more and more difficult to detect the high-quality forged money with a traditional money-counting machine. As the unique identification mark of the banknote, the crown word number is an important part in anti-counterfeiting measure. For cost reasons, the same crown word number are often be used in manufacturing a large number of counterfeit money. Based on this, if, the cash-counting machine can be used to recognize the crown word number in a real time and to detect the counterfeit money, the ability of cash counter counterfeit inspection can be improved greatly. At present, foreign countries have sophisticated cash counter with banknotes crown word number identification feature. At the same time, in this field, the domestic counter is still in the developing stage. This paper will introduce a crown word number recognition algorithm, which is an essential part of producing cash counter system with the crown word number recognition.In this paper, we propose a set of algorithms for the character image recognition in currency counting machine. Firstly, we analyzed three kinds of character recognition algorithms which are mainly used at present in detail. Secondly, we also selected the algorithm based on pattern matching method as the main algorithm in this paper. In order to increase the speed of the recognition algorithm, we propose a character image feature extraction method which calculates the number of character image connected regions. The recognition speed of the original algorithm is greatly increased by matching the character image based on classification using the number of character image connected regions as a feature. Moreover, in consideration of many similar characters in a character set is prone to errors when matching, we propose a character image extraction method which calculates the character image’s eight vertex positions. We improve the recognition accuracy by distinguish the similar characters in a character set using the feature of character image’s vertex positions. The details of our research are presented in this paper by the illustration and analysis of image rotation, image amplifying, image binarization, character segmentation in image preprocessing and image thinning, image compression in character recognition processing.At the end of this study, an experiment is conducted. The design and implementation of character recognition algorithms were tested in this experiment, which proves that, the algorithm can reach 900 or more per minute in processing banknote images, the character recognition accuracy can reach excess of 99% of the requirements. It is also confirmed that our character recognition algorithm is practical and efficient.