The Research Of ATM RMB Crown Code Recognition

With the sustainable development of Artificial Intelligence technology, the use of machines to replace human labor becomes inevitable in many fields. Along with the character recognition technology(OCR technology) matures, it becomes possible that people use the machines to replace human labor. RMB crown codes identification is an important application of OCR technology. The purpose of this paper is based on the characteristics of currency codes, to develop a system suitable for RMB crown recognition.Because of the images collected in this article are complete images of the currency and backgrounds of the collecting environment, rather than merely crown codes part of the image. To get the prefix zone image, this paper first obtained the tilt angle of the base and broadside of image by tilt angle measuring, followed by mapping and accurate positioning based on cross, for prefix regional localization and correction completed in the same time. Then we finish the exact location by the method of Pixel Frequency Changing. Gaussian filter is a good way of image smoothing and then we make the crown code binary image. In order to extract the codes to a single character effectively, this paper presents the method of based on the evaluation function of moving character frame. this method has high efficiency, good robustness, effectively solves breaking character and touching character segmentation problems. Use interpolation method in image normalization processing to make feature extraction easy. In the feature extraction, because of the satisfactory character image segmentation, it is easy to identify code. After the experimental comparison of crossing characteristics, boundary point feature, grid feature, combining algorithm speed requirements, we decided to adopt the grid feature which is easy to extract and the effect of identify can achieve the requirements of the subject. The methods of feature extraction is quick sensitive and easy. In the end, regarding the choice of a classifier, the paper uses a fast classification based on BP neural network classifier system and obtains good results.In the actual project, the algorithm reached the task requirements in terms of computational time and identify rate. In the tese of 1000 RMB crown codes using the training model of 720 samples, the identify rate reached 99.9%.