Dollar Bill Image Detection Of Critical Information And Number Recognition

With the global economic integration, accelerate the opening up of China, the U.S. dollar as world currency plays an irreplaceable role in the foreign exchange business, develop a U.S. dollar Counter which can quickly and accurately identify dollar's key word and record number plays an important role in our foreign exchange business.Firstly,This paper describes the dollar bill image pre-processing method, including gray-scale, smooth denoising, fisheye deformation correction, tilt correction, image enhancement. Pre-process the angle image and then detect the key information of the image, using the traditional template matching, PCA feature matching,8-direction gradient feature matching to do the experiment, from the correct rate, time-consuming point of view, 8-direction gradient feature matching method superior, and detect the real-time consuming, around 50ms/piece, meet the system's real-time requirements.Identify dollar's number, the first step is to locate the number and pre-process image,because dollar bill's face is too complex, and then locate number images precise in the X axis direction and Y axis direction respectively, obtain Number body image.Then combine global Ostu and local threshold methods to binary the number body image and use the vertical projection to segment single characters, and normalize single characters. For character recognition, feature select 8-direction gradient feature and PCA feature, classifier select the KNN classifier and BP neural network, using the PCA_KNN,8-direFea_KNN,PCA_BP three kinds of identification methods to do the experiment, beacuse number recognition as part of this system does not require real-time, so from the perspective of recognition rate, using 8-direFea_KNN to identify the bill number.Finally, realize image processing and recognition technology in the application of U.S. Counter, combined with the COMS image sensor MT9T001 and TMS320DM642 achieve a real-time embedded image acquisition and processing system, the system is stable, able to run off-line real-time, processing speed up to 800 piece/min, the correct rate of the dollar key information identification is about 99.5%, and the dollar bill character number identification rate is more than 92%.