基于变换的越南文手写文字图像识别模型

Vinh-Loi Ly, T. Doan, N. Ly
{"title":"基于变换的越南文手写文字图像识别模型","authors":"Vinh-Loi Ly, T. Doan, N. Ly","doi":"10.1109/NICS51282.2020.9335877","DOIUrl":null,"url":null,"abstract":"Handwritten text recognition plays an important role in transforming handwritten-based documents into digital data, which is necessary to intellectualize social management and production processes in the fourth industrial revolution. To overcome this challenge, several recent studies have assumed that each character appeared on the image is independent so that they could make predictions solely on the visual features. However, it leads to a lack of language characteristics because of the fact that the occurrence of a character is somehow related to the previous characters. Therefore, the attention mechanism between the text and the image to create the character predictions sequentially have outperformed the above method on the word level because it could make use of the context of the predicting word text. In this paper, which is inspired by the Transformer architecture in Neural Machine Translation tasks, we further proposed a model that exploits the dependencies between the last predicted character and the previously predicted characters based on the attention mechanism to translate from the word image to a word text. Our method has achieved the state-of-the-art result with 2.48% CER and 7.70% WER on the VNOnDB-word data set compared to similar works on the same data set.","PeriodicalId":308944,"journal":{"name":"2020 7th NAFOSTED Conference on Information and Computer Science (NICS)","volume":"48 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2020-11-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":"{\"title\":\"Transformer-based model for Vietnamese Handwritten Word Image Recognition\",\"authors\":\"Vinh-Loi Ly, T. Doan, N. Ly\",\"doi\":\"10.1109/NICS51282.2020.9335877\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Handwritten text recognition plays an important role in transforming handwritten-based documents into digital data, which is necessary to intellectualize social management and production processes in the fourth industrial revolution. To overcome this challenge, several recent studies have assumed that each character appeared on the image is independent so that they could make predictions solely on the visual features. However, it leads to a lack of language characteristics because of the fact that the occurrence of a character is somehow related to the previous characters. Therefore, the attention mechanism between the text and the image to create the character predictions sequentially have outperformed the above method on the word level because it could make use of the context of the predicting word text. In this paper, which is inspired by the Transformer architecture in Neural Machine Translation tasks, we further proposed a model that exploits the dependencies between the last predicted character and the previously predicted characters based on the attention mechanism to translate from the word image to a word text. Our method has achieved the state-of-the-art result with 2.48% CER and 7.70% WER on the VNOnDB-word data set compared to similar works on the same data set.\",\"PeriodicalId\":308944,\"journal\":{\"name\":\"2020 7th NAFOSTED Conference on Information and Computer Science (NICS)\",\"volume\":\"48 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2020-11-26\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"1\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2020 7th NAFOSTED Conference on Information and Computer Science (NICS)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/NICS51282.2020.9335877\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2020 7th NAFOSTED Conference on Information and Computer Science (NICS)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/NICS51282.2020.9335877","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 1

摘要

手写文本识别在将基于手写的文档转换为数字数据方面发挥着重要作用,这是第四次工业革命中社会管理和生产过程智能化所必需的。为了克服这一挑战,最近的几项研究假设图像上出现的每个字符都是独立的,这样他们就可以仅根据视觉特征做出预测。然而,由于一个字符的出现与前面的字符有某种联系,因此导致了语言特征的缺乏。因此,文本和图像之间的注意机制在单词层面上优于上述方法,因为它可以利用预测单词文本的上下文。在本文中,受神经机器翻译任务中的Transformer架构的启发,我们进一步提出了一种基于注意机制,利用最后预测的字符与之前预测的字符之间的依赖关系,从单词图像翻译到单词文本的模型。与同类方法相比,我们的方法在VNOnDB-word数据集上取得了2.48%的CER和7.70%的WER的最新结果。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
Transformer-based model for Vietnamese Handwritten Word Image Recognition
Handwritten text recognition plays an important role in transforming handwritten-based documents into digital data, which is necessary to intellectualize social management and production processes in the fourth industrial revolution. To overcome this challenge, several recent studies have assumed that each character appeared on the image is independent so that they could make predictions solely on the visual features. However, it leads to a lack of language characteristics because of the fact that the occurrence of a character is somehow related to the previous characters. Therefore, the attention mechanism between the text and the image to create the character predictions sequentially have outperformed the above method on the word level because it could make use of the context of the predicting word text. In this paper, which is inspired by the Transformer architecture in Neural Machine Translation tasks, we further proposed a model that exploits the dependencies between the last predicted character and the previously predicted characters based on the attention mechanism to translate from the word image to a word text. Our method has achieved the state-of-the-art result with 2.48% CER and 7.70% WER on the VNOnDB-word data set compared to similar works on the same data set.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
自引率
0.00%
发文量
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:604180095
Book学术官方微信