摘要
针对当前发票中被印章覆盖的字符识别准确率较低的问题,提出了一种基于像素灰度值检测的字符修补方法。首先,对带印章的一组字符进行三通道的分解得到方便后续处理的R通道图像;其次,根据R通道图像字符中间出现的断层这一情况,遍历图像中灰度值满足特定条件的像素,并通过一系列的阈值条件判断该像素是否属于断裂区域;最后,采用基于反向传播算法的多层感知分类器对已修补的字符进行识别。采用halcon自带的3个字符模板和对原始发票中的字符训练得到的模板分别对修补之后的字符进行识别,后者的识别效果不论在消耗的时间上还是准确率方面较前者都要好,准确率达95.38%。从识别消耗的时间和准确率来讲,提出的方法能够对含印章字符进行快速和准确的识别。
The work aims to propose a character repair method based on pixel gray value detection to solve the problem of low recognition of invoices contaminated by seals.Firstly,a set of characters with seal were decomposed into three channels to obtain image of R-channel for subsequent processing.Secondly,according to the fracture appearing in the middle of characters of R-channel image,pixels whose gray value meets certain conditions in the image were traversed,and a series of threshold conditions were used to judge whether the pixel belongs to the fracture region.Finally,a BP algorithm based Multi-Layer Perceptron classifier was used to recognize the patched characters.Three character templates provided by halcon and the template obtained from training of characters in the original invoice were used to recognize the characters after the repair respectively.The latter recognition effect is better than that of the former in terms of time consumption and accuracy and the accuracy rate reaches 95%.From aspects of time consumption and accuracy,the proposed method can quickly and accurately recognize the characters contaminated by seals.
作者
李镇锋
陈晓荣
管婷欣
郭蓉蓉
王晓龙
Li Zhenfeng;Chen Xiaorong;Guan Tingxin;Guo Rongrong;Wang Xiaolong(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《电子测量技术》
2020年第10期131-134,共4页
Electronic Measurement Technology
关键词
印章噪声
字符断裂
图像修复
字符识别
seal noise
character break
image restoration
character recognition
作者简介
李镇锋,硕士研究生,主要研究方向为图像处理与字符识别。E-mail:lizhenfeng1028@163.com;陈晓荣,博士,副教授,主要研究方向为图像处理、在线检测、信号与信息处理。E-mail:cxrsjtu@163.com
