Banknotes Serial Number Identification System Algorithm

Modernization of the financial system is a milestone in protecting the economic prosperity, and maintaining social harmony. In this article, a radial basis function networks based serial number recognition for RMB is introduced, during the researches of currency sorter.RMB Image data is acquired through CIS (Contact Image Sensor), converted into digital images by A/D converter, and then transfered to the DSP (Digital Signal Processer). On DSP chip, the image containing serial numbers is extracted from the RMB image, after most preprocessing steps and then transferred to PC, on which recognition is accomplished.Focus of this article is divided into image preprocessing and character recognition, including: noise reduction, image binarization, edge detection, distortion correction and serial number extraction, character segmentation, as well as pattern recognition. As to the individual steps of currency serial number recognition, commonly used algorithms and ideas are described in detail. And corresponding methods are introduced in this article according to the characteristics of RMB images and the character set of serial numbers.The radial basis function network based pattern recognition method is adopted, which generates 95% recognition rates, exceeding the counterpoints generated by the reported BP network [1], as well as with much fewer hidden nodes. In the follow-up work, higher recognition rates are achievable through further optimizations of training methods, while still maintaining small amounts of network nodes. Compared with commonly used methods of the currency serial number recognition, following steps are optimized:(1) Randomized Hough transformation is introduced in instead of commonly used linear fitting method, detecting RMB boundaries.(2) An affine transformation method is introduced in combination of individual rotation and tangent transformations. (3) A run-length-encoding based method is proposed to extract the serial numbers region(4) The connected component analysis method is adopted in character segmentation. In order to fulfill system's real-time requirement, the parallel optimization algorithms are also used.(5) A mixed feature extraction method is proposed in combination of the two-dimensional projection histogram and the Euler number.(6) The radial basis function networks are adopted as the pattern recognition method, and a network model is proposed for RMB serial number recognition.The adopted methods, such as the run-length encoding method , parallel-optimized connected component analysis method and the radial basis function network based pattern recognition method, as we know, haven't been reported in the domestic literatures on the application of paper currency recognition.The reported character recognition technology could be widely used in the manufacture, maintainances and management of import equipments and weapons, such as motor vehicles, guns, tanks and so on.