基于特征的BP神经网络人民币号码识别系统

A RMB Number Recognition System of BP Neural Network Based on Features

中文关键词:  人民币号码  特征提取  BP神经网络  训练  像素  预处理

英文关键词:RMB number  feature extraction  BP neural network  training  pixel  preprocess

中文摘要:

      介绍了一种简单、准确、快速的人民币号码识别系统。采集到的人民币号码图像经过预处理、特征提取和神经网络训练识别几个过程识别出号码串。特征提取采用的是水平方向5行逐像素特征提取法，该方法的优点是提取相对较少的特征点即可反映丰富的信息量。实验结果表明，提出的特征提取方法可以使BP神经网络很快收敛，训练效果也较好，该识别系统能高效准确地提取出人民币号码串，识别率达到95%以上，平均识别时间为35 ms，适应性也较强。

英文摘要:

      A simple,accurate and fast RMB number recognition system is introduced.Acquired RMB number images are identified in number by the processes of preprocessing,feature extraction and neural network training recognition.Pixel by pixel method in five rows of horizontal direction is presented for extracting feature.There are advantages of this method that extracted feature points are less and information contents are abundant.Experimental results show that the BP neural network can converge fast and the training effect is perfect in the presented feature extraction method.RMB numbers can be extracted efficiently and accurately by this system.The correct recognition rate is above 95%,the researched system spends 35 ms in recognizing an image every time and is highly adaptable.