Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (9): 207-211.DOI: 10.3778/j.issn.1002-8331.2002-0015

Previous Articles     Next Articles

Improved Handwritten Date Recognition in Scanned Documents Combined with LeNet-5

ZHANG Cheng, DAI Junfeng, XIONG Wenxin   

  1. 1.State Grid Hubei Information & Telecommunication Company Limited, Wuhan 430077, China
    2.School of Electronic Information, Wuhan University, Wuhan 430072, China
  • Online:2021-05-01 Published:2021-04-29

融合LeNet-5改进的扫描文档手写日期识别

张成,戴俊峰,熊闻心   

  1. 1.国网湖北省电力有限公司 信息通信公司,武汉 430077
    2.武汉大学 电子信息学院,武汉 430072

Abstract:

This paper studies the application of LeNet-5 in handwritten date characters recognition in scanned documents. As various noises will appear in the process of document scanning, especially light and color interference, using LeNet-5 algorithm directly can not get good results. This article firstly gets the location and division of the particular character to be recognized in the whole document while the divided character image is processed by denoising, graying and binarization. For the next step, character image is segmented into a single character and then the article realizes the recognition of handwritten date characters on the basis of LeNet-5 network combined with model matching method. By comparing the recognition effect under different combination of parameters and adjusting the parameters to improve model performance for practical objects, an algorithm that can achieve a better recognition effect for handwritten date character set is realized. Experimental results show that the algorithm is effective and can be applied in practical engineering.

Key words: deep learning, convolutional neural network, LeNet-5, character recognition

摘要:

研究LeNet-5在扫描文档中手写体日期字符识别的应用,由于文档扫描的过程中会引入各种噪声,特别是光照和颜色干扰,直接使用LeNet-5算法不能取得较好效果。先在整份文档中对特定待识别字符的进行定位和划分,并对划分出的字符图像进行去噪、灰度化和二值化处理等预处理,接着将字符图像分割成一个个单个字符,然后在LeNet-5网络基础上结合模型匹配法实现对手写体日期字符的识别。分析在不同参数组合下的识别效果,调整算法模型参数有效地提升了模型对于实际对象的性能,实现出一种能够对手写体日期字符集实现较好识别效果的算法。实验结果表明了算法的有效性,并应用于具体工程实践。

关键词: 深度学习, 卷积神经网络, LeNet-5, 字符识别