适合特定水表表型的字符识别方法  

Character recognition algorithm adapt to specific kind of water meter

作　　者：李尧[1,2,3] 芮小平[3] 

机构地区：[1]内江师范学院计算机与信息科学系,四川内江641112 [2]内江师范学院网络应用项目开发重点实验室,四川内江641112 [3]中国科学院研究生院资环学院,北京100049

出　　处：《计算机工程与设计》2009年第11期2772-2774,2781,共4页Computer Engineering and Design

基　　金：四川省应用基础重点基金项目(07JY29-124);四川省教育厅重点基金项目(2006A145)

摘　　要：根据特定水表走字的特点,提出了一种基于模板和神经网络的水表数字字符识别的方法。该方法利用特定表型中待识别区域宽高比例相同的特点,提出用特征模板缩放法来进行待识别区域的定位和字符分割。采用三灰度值加权系数进行模板匹配,提高了特征的利用率;采用自适应学习的BP神经网络训练全字符和半字符样本,用户直接使用训练好的神经网络联结权值进行字符识别。结果表明,充分利用待识别区域的特征有助于提高识别区域定位和字符分割的准确性,在此基础上,采用经典的识别算法能够取得较好的效果。An algorithm to recognize the numerical characters for a given kind of water meter is brought forward. Because of the same scales of the width and height of the target area, this algorithm makes a feature pattern which has three gray scales according to the target area and zooms the feature pattern to fit for segment. These three gray scales divide the whole 256 gray scales of our image into 7 types and the authors define the different weights for these 7 kinds of gray scales when making feature pattern matching. Adaptive learning BP neural networks is used to train the samples of 10 whole characters and 10 half characters, the software recognize the character based on the trained BP neural network. The results show that full using of the features of the area to be recognized can improved the accuracy of locating the area and segmenting of the characters, the classical recognition algorithm can obtain a good effect based on these methods.

关 键 词：字符识别 定位 BP神经网络 模板匹配 图像采集 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]