| With the rapid development of computer technology, automatic and intelligent systems are used more and more widely in all walks of life. Intelligent recognition system for paper currency is very popular in the bank for it can automatically identify paper currency’s authenticity, value and serial number. Paper currency’s classification is also available according to its quality with this kind of system. However, this powerful system is always large-scale, expensive and not energe-saving. Therefore, portable energe-saving and cost-effective embedded recog-nition system for paper currency is more and more popular in medium-sized enterprises. This thesis researches and designs a cost-effective embedded recognition system for paper currency on this condition.In this thesis, many parts of the existing algorithm have been improved. Propose a new method of image binarization which takes the global threshold and local distribution into a combination. Use edge image binarization and section image binarization to meet the feature of paper currency images. A new method of serial numbers location based on template matching is proposed to solve the problems of inefficiency and the insufficien accuracy which the old algorithm has. Adjust methods of denoising, boundary location and serial numbers segmentation and so on. Now the accuracy of boundary location is improved from 93% to 100%. The accuracy of characters’segmentation in serial number is also lagerly improved. This thesis chooses the cost-effective DSP DM642 as the processing unit of our embedded system, finishes the program porting from Windows to DSP platform and rewrites Vector class and parts of OpenCV functions using C language. In order to improve processing speed, multi-level optimization was conducted on this platform, including C code optimization and DSP Assembly optimization. After optimization, the processing time per frame of value recognition and serial numbers recognition both dropped from more than 1000ms to 20ms or less and meet the need of realtime-processing. |
