摘要
纸币面向识别是纸币识别的基础,传统的纸币面向识别方法是人工提取特征,对于污损严重的纸币图像识别效率不高。针对传统方法的缺点,提出一种针对纸币图像的预处理方法。使用基于改进的BP神经网络的纸币面向识别方法,采用对纸币图像分块取平均值的方法提取特征,用量化共轭梯度法进行神经网络的训练,并且在TMS320DM648上进行实现。实验结果表明,这种方法完成纸币图像预处理和面向识别的时间不超过25 ms,准确率高于99%,具有计算量小、识别结果正确率高等优点。
Banknote recognition in regard to surface and direction is the basis of banknote recognition. Traditional banknotes recognition method extracts the features of surface and direction of banknote manually, which has low recognition rate on seriously defaced banknote images. For the shortcoming of traditional methods, we proposed a new preprocessing method for banknote images. We used the improved BP neural network-based banknotes surface and direction recognition method, extracted features by the method of taking the mean of banknote image blocks, and used scaled conjugate gradient method to train neural networks, then implemented the new method on TMS320DM648. Experimental results showed that this method completed the banknote image preprocessing and surface and direction recognition within 25 ms in time and its accuracy was higher than 99%, it had the advantages of small computation load and high recognition accuracy rate.
出处
《计算机应用与软件》
CSCD
2015年第11期176-179,共4页
Computer Applications and Software
关键词
纸币面向
图像预处理
神经网络
量化共轭梯度法
Banknote surface and direction
Image preprocess
Neural network
Scaled conjugate gradient （SCG）
作者简介
刘艳萍，教授，主研领域：DSP技术及FP-GA技术。
杜秋晨，硕士生。
张进东，工程师。
