摘要
手写体数字字符串识别常用于邮件自动分拣、银行票据和财务报表的录入中,针对其分割识别算法复杂度较高、准确率较低的问题,提出一种多分类器下无分割手写数字字符串识别算法。该算法的核心是采用四个分类器实现粘连字符串的无分割识别;将残差结构应用于Le Net-5网络,以增加网络深度,提高识别准确率,加快收敛速度;使用动态选择策略,以避免长度分类器误分类对识别结果的影响。实验结果表明,在NIST SD19一位数字和Synthetic数据集训练网络下,使用NIST SD19上长度为2、3、4、5、6的字符串验证网络,其识别准确率分别为99.3%、98.5%、98.1%、96.6%和97.2%。
Handwritten numeral string recognition is often used in automatic mail sorting,bank bills and financial statements input.To overcome the high complexity and low accuracy of its segmentation and recognition algorithm,this paper proposed a non-segmentation handwritten numeral string recognition algorithm based on multi-classifier.The core of the algorithm used four classifiers to realize non-segmented recognition of sticky strings,applied residual structure to LeNet-5 network to increase network depth,improved recognition accuracy and speeded up convergence,and used dynamic selection strategy to avoid the impact of false classification of length classifiers on recognition results.The experimental results show that the recognition accuracy of NIST SD19 is 99.3%,98.5%,98.1%,96.6%and 97.2%respectively by using a 2,3,4,5,6-length string validation network on NIST SD19.
作者
任晓奎
丁鑫
陶志勇
何欣键
Ren Xiaokui;Ding Xin;Tao Zhiyong;He Xinjian(School of Electronics&Information Engineering,Liaoning Technical University,Huludao Liaoning 125105,China;Fuxin Lixing Technology Co,Ltd,Fuxin Liaoning 123000,China;State Grid Aba Electric Power Company Limited,Aba Sichuan 623200,China)
出处
《计算机应用研究》
CSCD
北大核心
2020年第7期2222-2226,共5页
Application Research of Computers
基金
辽宁省博士启动基金资助项目(20170520098)
辽宁省自然科学基金资助项目(2015020100)
辽宁省普通高等教育本科教学改革研究项目(551610001095)
辽宁省教育厅一般项目(LJ2017QL013)。
关键词
图像处理
手写数字字符串识别
多分类器
无分割
动态选择
image processing
handwritten digital string recognition
multi-classifier
no segmentation
dynamic selection
