Design Of Paper Currency Recognition System Based On DSP

With the rapid development of computer technology, financial field reaches the stage of automation, and use of paper currency classifier in financial field is wider and wider, making banknote image recognition become a popular topic and wide promising.Banknote image real-time processing and recognition is the core technology of paper currency classify system, function and research situation of paper currency classifier are summarized at first, software architecture and hardware architecture of classify system are presented in general, function and structure of system are described. Then, research is achieved based on existing paper currency preprocessing algorithm, including image denoising, image edge extraction and skew correction. Value and orientation are the most prominent features of paper currency, value recognition is based on width of banknote image and orientation recognition is based on neural network, which makes use of global image region, with the advantage of high stability and little impact of noise. Considering further banknote recognition, methods of recognizing serial number, condition and authenticity are proposed. In serial number recognition, use projection method to segment serial number and use template matching method to recognize. Classification of new and used bills is based on grayscale comparison and authenticity recognition is based on infrared image of banknote, using grayscale ratio to distinguish between true and false.High performance TMS32DM648 of TI cooperation C6000 series is chosen as processing unit of classifier system, transplanting recognition algorithm to this platform. In order to improve code efficiency, DSP code optimization is achieved, including C code optimization and assembly optimization on critical part of system, which shorten the time for value and orientation recognition into 10 ms and for serial recognition into 20 ms, meeting the requirement of real-time performance.