Transformer-based architectures show excellent results on the task of handwritten text recognition, becoming the standard architecture for modern datasets. However, they require a significant amount of annotated data to achieve competitive results. They typically rely on synthetic data to solve this problem. Historical handwritten text recognition represents a challenging task due to degradations, specific handwritings for which few examples are available and ancient languages that vary over time. These limitations also make it difficult to generate realistic synthetic data. Given sufficient and appropriate data, Transformer-based architectures could alleviate these concerns, thanks to their ability to have a global view of textual images and their language modeling capabilities. In this paper, we propose the use of a lightweight Transformer model to tackle the task of historical handwritten text recognition. To train the architecture, we introduce realistic looking synthetic data reproducing the style of historical handwritings. We present a specific strategy, both for training and prediction, to deal with historical documents, where only a limited amount of training data are available. We evaluate our approach on the ICFHR 2018 READ dataset which is dedicated to handwriting recognition in specific historical documents. The results show that our Transformer-based approach is able to outperform existing methods.
{"title":"Training transformer architectures on few annotated data: an application to historical handwritten text recognition","authors":"Killian Barrere, Yann Soullard, Aurélie Lemaitre, Bertrand Coüasnon","doi":"10.1007/s10032-023-00459-2","DOIUrl":"https://doi.org/10.1007/s10032-023-00459-2","url":null,"abstract":"<p>Transformer-based architectures show excellent results on the task of handwritten text recognition, becoming the standard architecture for modern datasets. However, they require a significant amount of annotated data to achieve competitive results. They typically rely on synthetic data to solve this problem. Historical handwritten text recognition represents a challenging task due to degradations, specific handwritings for which few examples are available and ancient languages that vary over time. These limitations also make it difficult to generate realistic synthetic data. Given sufficient and appropriate data, Transformer-based architectures could alleviate these concerns, thanks to their ability to have a global view of textual images and their language modeling capabilities. In this paper, we propose the use of a lightweight Transformer model to tackle the task of historical handwritten text recognition. To train the architecture, we introduce realistic looking synthetic data reproducing the style of historical handwritings. We present a specific strategy, both for training and prediction, to deal with historical documents, where only a limited amount of training data are available. We evaluate our approach on the ICFHR 2018 READ dataset which is dedicated to handwriting recognition in specific historical documents. The results show that our Transformer-based approach is able to outperform existing methods.</p>","PeriodicalId":50277,"journal":{"name":"International Journal on Document Analysis and Recognition","volume":"26 1","pages":""},"PeriodicalIF":2.3,"publicationDate":"2024-01-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"139580505","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-12-08DOI: 10.1007/s10032-023-00457-4
Tauseef Iftikhar, Nazar Khan
We tackle a novel problem of detecting background grids in hand-drawn cadastral maps. Grid extraction is necessary for accessing and contextualizing the actual map content. The problem is challenging since the background grid is the bottommost map layer that is severely occluded by subsequent map layers. We present a novel automatic method for robust, bottom-up extraction of background grid structures in historical cadastral maps. The proposed algorithm extracts grid structures under significant occlusion, missing information, and noise by iteratively providing an increasingly refined estimate of the grid structure. The key idea is to exploit periodicity of background grid lines to corroborate the existence of each other. We also present an automatic scheme for determining the ‘gridness’ of any detected grid so that the proposed method self-evaluates its result as being good or poor without using ground truth. We present empirical evidence to show that the proposed gridness measure is a good indicator of quality. On a dataset of 268 historical cadastral maps with resolution (1424times 2136) pixels, the proposed method detects grids in 247 images yielding an average root-mean-square error (RMSE) of 5.0 pixels and average intersection over union (IoU) of 0.990. On grids self-evaluated as being good, we report average RMSE of 4.39 pixels and average IoU of 0.991. To compare with the proposed bottom-up approach, we also develop three increasingly sophisticated top-down algorithms based on RANSAC-based model fitting. Experimental results show that our bottom-up algorithm yields better results than the top-down algorithms. We also demonstrate that using detected background grids for stitching different maps is visually better than both manual and SURF-based stitching.
{"title":"Background grid extraction from historical hand-drawn cadastral maps","authors":"Tauseef Iftikhar, Nazar Khan","doi":"10.1007/s10032-023-00457-4","DOIUrl":"https://doi.org/10.1007/s10032-023-00457-4","url":null,"abstract":"<p>We tackle a novel problem of detecting background grids in hand-drawn cadastral maps. Grid extraction is necessary for accessing and contextualizing the actual map content. The problem is challenging since the background grid is the bottommost map layer that is severely occluded by subsequent map layers. We present a novel automatic method for robust, bottom-up extraction of background grid structures in historical cadastral maps. The proposed algorithm extracts grid structures under significant occlusion, missing information, and noise by iteratively providing an increasingly refined estimate of the grid structure. The key idea is to exploit periodicity of background grid lines to corroborate the existence of each other. We also present an automatic scheme for determining the ‘gridness’ of any detected grid so that the proposed method self-evaluates its result as being good or poor without using ground truth. We present empirical evidence to show that the proposed gridness measure is a good indicator of quality. On a dataset of 268 historical cadastral maps with resolution <span>(1424times 2136)</span> pixels, the proposed method detects grids in 247 images yielding an average root-mean-square error (RMSE) of 5.0 pixels and average intersection over union (IoU) of 0.990. On grids self-evaluated as being good, we report average RMSE of 4.39 pixels and average IoU of 0.991. To compare with the proposed bottom-up approach, we also develop three increasingly sophisticated top-down algorithms based on RANSAC-based model fitting. Experimental results show that our bottom-up algorithm yields better results than the top-down algorithms. We also demonstrate that using detected background grids for stitching different maps is visually better than both manual and SURF-based stitching.</p>","PeriodicalId":50277,"journal":{"name":"International Journal on Document Analysis and Recognition","volume":"21 1","pages":""},"PeriodicalIF":2.3,"publicationDate":"2023-12-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"138556858","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
StubbeAndrea, RinglstetterChristoph, U. SchulzKlaus
Given a specific information need, documents of the wrong genre can be considered as noise. From this perspective, genre classification helps to separate relevant documents from noise. Orthographic...
{"title":"Genre as noise","authors":"StubbeAndrea, RinglstetterChristoph, U. SchulzKlaus","doi":"10.2307/j.ctv125jncf.8","DOIUrl":"https://doi.org/10.2307/j.ctv125jncf.8","url":null,"abstract":"Given a specific information need, documents of the wrong genre can be considered as noise. From this perspective, genre classification helps to separate relevant documents from noise. Orthographic...","PeriodicalId":50277,"journal":{"name":"International Journal on Document Analysis and Recognition","volume":"15 1","pages":""},"PeriodicalIF":2.3,"publicationDate":"2007-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84556320","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
There is a huge amount of historical documents in libraries and in various National Archives that have not been exploited electronically. Although automatic reading of complete pages remains, in mo...
在图书馆和各个国家档案馆中有大量的历史文献没有被电子利用。虽然自动读取完整的页面仍然存在，在…
{"title":"Text line segmentation of historical documents: a survey","authors":"Likforman-SulemLaurence, ZahourAbderrazak, TaconetBruno","doi":"10.5555/1237480.1237483","DOIUrl":"https://doi.org/10.5555/1237480.1237483","url":null,"abstract":"There is a huge amount of historical documents in libraries and in various National Archives that have not been exploited electronically. Although automatic reading of complete pages remains, in mo...","PeriodicalId":50277,"journal":{"name":"International Journal on Document Analysis and Recognition","volume":"43 1","pages":""},"PeriodicalIF":2.3,"publicationDate":"2007-04-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85190118","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
There is a huge amount of historical documents in libraries and in various National Archives that have not been exploited electronically. Although automatic reading of complete pages remains, in mo...
在图书馆和各个国家档案馆中有大量的历史文献没有被电子利用。虽然自动读取完整的页面仍然存在，在…
{"title":"Text line segmentation of historical documents","authors":"Likforman-SulemLaurence, ZahourAbderrazak, TaconetBruno","doi":"10.5555/2722890.2723025","DOIUrl":"https://doi.org/10.5555/2722890.2723025","url":null,"abstract":"There is a huge amount of historical documents in libraries and in various National Archives that have not been exploited electronically. Although automatic reading of complete pages remains, in mo...","PeriodicalId":50277,"journal":{"name":"International Journal on Document Analysis and Recognition","volume":"77 1","pages":""},"PeriodicalIF":2.3,"publicationDate":"2007-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85564508","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2003-04-01DOI: 10.1007/s10032-002-0085-5
A. Britto, R. Sabourin, Flávio Bortolozzi
{"title":"The recognition of handwritten numeral strings using a two-stage HMM-based method","authors":"A. Britto, R. Sabourin, Flávio Bortolozzi","doi":"10.1007/s10032-002-0085-5","DOIUrl":"https://doi.org/10.1007/s10032-002-0085-5","url":null,"abstract":"","PeriodicalId":50277,"journal":{"name":"International Journal on Document Analysis and Recognition","volume":"7 1","pages":"102-117"},"PeriodicalIF":2.3,"publicationDate":"2003-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89695619","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2003-04-01DOI: 10.1007/s10032-002-0098-0
Lixin Fan, Liying Fan, C. Tan
{"title":"Adaptive image-smoothing using a coplanar matrix and its application to document image binarization","authors":"Lixin Fan, Liying Fan, C. Tan","doi":"10.1007/s10032-002-0098-0","DOIUrl":"https://doi.org/10.1007/s10032-002-0098-0","url":null,"abstract":"","PeriodicalId":50277,"journal":{"name":"International Journal on Document Analysis and Recognition","volume":"262 1","pages":"88-101"},"PeriodicalIF":2.3,"publicationDate":"2003-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"82471177","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2003-04-01DOI: 10.1007/s10032-002-0093-5
A. Spitz, K. Tombre
{"title":"Special issue – selected papers from the ICDAR'01 conference","authors":"A. Spitz, K. Tombre","doi":"10.1007/s10032-002-0093-5","DOIUrl":"https://doi.org/10.1007/s10032-002-0093-5","url":null,"abstract":"","PeriodicalId":50277,"journal":{"name":"International Journal on Document Analysis and Recognition","volume":"25 1","pages":"87"},"PeriodicalIF":2.3,"publicationDate":"2003-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"82836000","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2003-04-01DOI: 10.1007/s10032-002-0091-7
David J. Crandall, Sameer Kiran Antani, R. Kasturi
{"title":"Extraction of special effects caption text events from digital video","authors":"David J. Crandall, Sameer Kiran Antani, R. Kasturi","doi":"10.1007/s10032-002-0091-7","DOIUrl":"https://doi.org/10.1007/s10032-002-0091-7","url":null,"abstract":"","PeriodicalId":50277,"journal":{"name":"International Journal on Document Analysis and Recognition","volume":"1 1","pages":"138-157"},"PeriodicalIF":2.3,"publicationDate":"2003-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"77103825","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号