Research On Paper Currency Numbers Recognition Algorithm

Banknote is the blood of finance and indispensable in the world nowadays. By using the technology of digital image processing and pattern recognition, this article does research work on paper currency recognition and designs algorithms of character recognition towards dollar which is the most influential currency in modern world.Paper currency recognition mainly includes parts of including paper currency image pre-processing, incline rectification, serial number segments, single character segments and character recognition. The paper currency recognition system in this article achieved the whole process from image acquisition to recognition result. And image pre-processing and methods of character recognition are studied mainly. A new method suitable for real-time implement by ARM is proposed and is simulated on PC.Firstly, paper currency image pre-processing is done, which includes image denoising, binarization, smoothing and image filtering of the binarized image. This thesis researches three methods of image binarazation systematically, i.e. the bimodal method, iterative algorithm and the Otsu method, among which the iterative algorithm gets the most satisfied result in the recognition system. After binarization there still exist many noise blots in the image, so a new Unger smoothing algorithm is proposed and obtains effective result.Secondly, key algorithms of paper currency recognition are studied, which include image correction and detecting, serial number area extraction, single character segmentation and thinning algorithms. A new slope angle detection method is proposed and rotating the image simultaneously. The location of paper currency can be precisely got by using the vertical and horizontal projection. Then using row and column segmentation is used to get serial number from image. At last a new method is used to segment single character which will be used as a recognition unitLastly, On the basis of studing the traditional method of character recognition, this Thesis proposes a new method based on structure character and then create a recognition tree to realize the serial number recognition work.