基于“字体循环gan”的生字数据增强在字符识别中的应用

2021 IEEE International Conference on Emergency Science and Information Technology (ICESIT) Pub Date : 2021-11-22 DOI:10.1109/ICESIT53460.2021.9696657

Kai Wu, Dingjiang Yan, Hongcheng Liao, Xiang Zhang, Q. Huang, Qian Zhang, Min Fu

{"title":"基于“字体循环gan”的生字数据增强在字符识别中的应用","authors":"Kai Wu, Dingjiang Yan, Hongcheng Liao, Xiang Zhang, Q. Huang, Qian Zhang, Min Fu","doi":"10.1109/ICESIT53460.2021.9696657","DOIUrl":null,"url":null,"abstract":"Given the low efficiency of massive image data processing in actual business departments, image character recognition OCR (optical character recognition) based on deep learning neural networks is not high. Based on the above background, a text image data enhancement method based on Font-CycleGan is proposed to improve text recognition accuracy. Our goal is to generate the data of specific fonts for the original hidden word text samples through the font library and then enhance the data of the hidden word text through CycleGan to enrich the database of the hidden word text to improve the performance of the classifier and the accuracy of text recognition. We evaluate and compare the impact of traditional character recognition and font-cyclgan based data to enhance character recognition accuracy. The results show that the improvement effect of this method is more significant.","PeriodicalId":164745,"journal":{"name":"2021 IEEE International Conference on Emergency Science and Information Technology (ICESIT)","volume":"291 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2021-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Application of Data Augmentation of Rare Words Based on “Font-CycleGan” in Character Recognition\",\"authors\":\"Kai Wu, Dingjiang Yan, Hongcheng Liao, Xiang Zhang, Q. Huang, Qian Zhang, Min Fu\",\"doi\":\"10.1109/ICESIT53460.2021.9696657\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Given the low efficiency of massive image data processing in actual business departments, image character recognition OCR (optical character recognition) based on deep learning neural networks is not high. Based on the above background, a text image data enhancement method based on Font-CycleGan is proposed to improve text recognition accuracy. Our goal is to generate the data of specific fonts for the original hidden word text samples through the font library and then enhance the data of the hidden word text through CycleGan to enrich the database of the hidden word text to improve the performance of the classifier and the accuracy of text recognition. We evaluate and compare the impact of traditional character recognition and font-cyclgan based data to enhance character recognition accuracy. The results show that the improvement effect of this method is more significant.\",\"PeriodicalId\":164745,\"journal\":{\"name\":\"2021 IEEE International Conference on Emergency Science and Information Technology (ICESIT)\",\"volume\":\"291 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2021-11-22\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2021 IEEE International Conference on Emergency Science and Information Technology (ICESIT)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/ICESIT53460.2021.9696657\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2021 IEEE International Conference on Emergency Science and Information Technology (ICESIT)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ICESIT53460.2021.9696657","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}

引用次数: 0

摘要

鉴于实际业务部门对海量图像数据处理效率较低，基于深度学习神经网络的图像字符识别OCR(光学字符识别)的要求不高。基于上述背景，本文提出了一种基于Font-CycleGan的文本图像数据增强方法，以提高文本识别的准确率。我们的目标是通过字体库对原始隐藏词文本样本生成特定字体的数据，然后通过CycleGan对隐藏词文本数据进行增强，丰富隐藏词文本数据库，从而提高分类器的性能和文本识别的准确率。我们评估和比较了传统字符识别和基于字体的数据对提高字符识别精度的影响。结果表明，该方法的改进效果更为显著。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

查看原文本刊更多论文

Application of Data Augmentation of Rare Words Based on “Font-CycleGan” in Character Recognition

Given the low efficiency of massive image data processing in actual business departments, image character recognition OCR (optical character recognition) based on deep learning neural networks is not high. Based on the above background, a text image data enhancement method based on Font-CycleGan is proposed to improve text recognition accuracy. Our goal is to generate the data of specific fonts for the original hidden word text samples through the font library and then enhance the data of the hidden word text through CycleGan to enrich the database of the hidden word text to improve the performance of the classifier and the accuracy of text recognition. We evaluate and compare the impact of traditional character recognition and font-cyclgan based data to enhance character recognition accuracy. The results show that the improvement effect of this method is more significant.

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

2021 IEEE International Conference on Emergency Science and Information Technology (ICESIT)

自引率

0.00%

发文量