2009 International Conference on Industrial Mechatronics and Automation 978-1-4244-3818-1/09/$25.00 ©2009 IEEE ICIMA 2009 Recognition of the Numbers of Numerical Civilian Instrumentations Based on BP Neural Network Qiushi BAI, Yunzhou ZHANG, Jiyuan TAN, Limeng ZHAO,Zixin QI College of Information Science and Engineering Northeastern University Shenyang, China Xiaobai3979@hotmail.com Abstract—With the rapid development of intelligent building, the requirement of automatic number identification of civilian instrumentations is increasingly urgent. This article uses iterative global threshold to binarize the images and then adopts projection method to locate the target regions and divide the numbers. The Back- Propagation Neural Network is used to recognize the numbers. The result indicates that the recognition rate is above 98%. Keywords-number recognition; civilian instrumentation; BP; neural network I. INTRODUCTION Recently, the recognition of numerical meters based on images has been applied in many automation fields and pattern recognition fields [1]. Different from earlier recognition of industrial numerical instrumentations, this article focuses on the recognition of the numerical civilian instrumentations applied widely in homes and campuses, including water meters, ammeters and gas meters. The conditions for the recognition of these meters are better than the industrial instrumentations, so it’s easier to apply algorithms to ensure the accuracy. What’s more, with the rapid development of intelligent building, the fact is that the needs of automatic reading of civilian instrumentations are increasingly urgent. For these instrumentations, this article uses the Back Propagation Neural Network to recognize the images of the instrumentations, and figure out the numbers finally. The result indicates that the ratio of the recognized numbers is more than 98%. II. THE RECOGNITION PROCESS As for the recognition process, it is made up of four sections as below: a) The pretreatment of the images. b) The locating of the target regions. c) The partition of the characters. d) The recognition of the characters. Figure 1. The process of the pretreatment Figure 1 shows the process of pretreatment. A. The Enhancement by Histogram Enhancement by histogram[2] is to modify the gray histogram of the images through gray-scale transformation. The object is to enhance contrast, to expanse the difference between the characters of different objects in the images, and to highlight the useful information. The usual way is Histogram equalization, which can distribute the origin histogram approximately evenly in the whole dynamic gray-scale and enhance the brightness of the images. The effects before and after the enhancement by Histogram are shown in Figure 2. Figure 2. The effects before and after the enhancement by histogram B. Median Filter When the practical system captures images, there is some noise because of the camera lens. The images enhanced by histogram can also be polluted because of the decrease of the information. Most of the usual ways to eliminate noise can eliminate isolated points but can also damage the edges of objects in the images. As for the recognition of the number characters on numerical instrumentations, the edges are important. So the median filter is applied here to eliminate the isolated points and single-line noises, the detail information of the edges protected. Enhancement by histogram Median Filtering Binarizing Input images 105 Authorized licensed use limited to: Northeastern University. Downloaded on June 24,2010 at 04:12:31 UTC from IEEE Xplore. Restrictions apply.