摘要
为了对非固定样式的手写表格文档进行批量识别处理,实现自动录入功能,首先通过空表识别生成单元格信息,分析版面结构;其次对图片进行去噪、倾斜校正、二值化等处理,对手写内容进行分割;最后搭建识别手写字符的卷积神经网络。实验结果表明,最终实现的识别系统能对不同格式的手写表格进行识别并生成数据格式文件。基于空表识别得到单元格信息的手写表格识别系统能对不同样式的表格进行批量识别处理,且通过CNN搭建识别模型,手写汉字也能被识别,使系统通用性更好,便于应用扩展。
In order to carry on the batch recognition processing to the non-fixed style handwritten documents and realize the function of automatic input,at first,the cell structure is generated by identifying the empty table,and the layout structure is analyzed.Then, the image is processed by denoising,skew correction,binarization,and the handwritten content is segmented.Finally,a convolutional neural network for handwritten character recognition is built.The experimental results show that the recognition system can recognize different handwritten forms and generate data format files.The handwritten form recognition system based on recognizing blank table and getting cell information can recognize different forms in batches,and the recognition model is set up by CNN,so handwritten Chinese characters can be recognized,which increases the generality of the system and easiness for application.
作者
李若月
钱强
张瀚文
方利堃
LI Ruo-yue;QIAN Qiang;ZHANG Han-wen;FANG Li-kun(School of Information Science and Technology,Southwest Jiaotong University,Chengdu 611756,China)
出处
《软件导刊》
2019年第5期17-20,26,共5页
Software Guide
基金
四川省大学生创新创业训练计划项目(201810613056)
关键词
手写汉字识别
表格识别
卷积神经网络
识别系统
handwritten Chinese character recognition
form recognition
convolutional neural network
recognition system
