基于卷积神经网络的批量发票识别系统研究

Batch Invoice Recognition System Based on Convolutional Neural Network(CNN)

针对增值税发票设计了一款基于卷积神经网络的批量发票识别系统。系统采用扫描仪对多张发票进行图像采集,然后对发票图像进行归一化处理,信息定位,投影算法的字符切割,再用CNN卷积神经网络进行字符识别。系统再根据信息间的相互联系,自动判断信息识别是否有误,可进行人工修改,提高系统的容错性。最后,将正确的信息自动记录到Excel表中相应的位置。对系统性能进行测试,50张随机发票测试结果:47张发票识别完全正确,3张发票的部分字符识别错误,系统能自动检测出识别错误并人工修改,使50张发票信息记录完全正确。

This paper designs a batch invoice recognition system based on convolution neural network for VAT(Value added tax) invoices.The scanner is used to collect images of invoices.First of all,the images is normalized to have the same size.Then.the identified information is located,and characters are cut based on projection algorithm.Last but not least,the extracted characters are recognized by convolution neural network.The system can automatically judge whether the information recognition is wrong or not according to the interrelation of the information.Besides,the wrong identified information can be modified manually to improve the fault tolerance of the system.Finally,the correct information is automatically recorded to the corresponding location in the Excel table.