In this paper, we propose a novel method for Automatic Text Recognition (ATR) on early printed books. Our approach signicantly reduces the Character Error Rates (CERs) for book-specic training when only a few lines of Ground Truth (GT) are available and considerably outperforms previous methods. An ensemble of models is trained simultaneously by optimising each one independently but also with respect
to a fused output obtained by averaging the individual condence matrices. Various experiments on ve early printed books show that this approach already outperforms the current state-of-the-art by up to 20% and 10% on average. Replacing the averaging of the condence matrices during prediction with a condence-based voting boosts our results by an additional 8% leading to a total average improvement of about 17%.
%0 Journal Article
%1 wick2021onemodel
%A Wick, Christoph
%A Reul, Christian
%B Proceedings of the 16th International Conference on Document Analysis and Recognition ICDAR 2021
%D 2021
%K myown myown:selected
%T One-Model Ensemble-Learning for Text Recognition of Historical Printings
%U https://link.springer.com/chapter/10.1007/978-3-030-86549-8_25
%X In this paper, we propose a novel method for Automatic Text Recognition (ATR) on early printed books. Our approach signicantly reduces the Character Error Rates (CERs) for book-specic training when only a few lines of Ground Truth (GT) are available and considerably outperforms previous methods. An ensemble of models is trained simultaneously by optimising each one independently but also with respect
to a fused output obtained by averaging the individual condence matrices. Various experiments on ve early printed books show that this approach already outperforms the current state-of-the-art by up to 20% and 10% on average. Replacing the averaging of the condence matrices during prediction with a condence-based voting boosts our results by an additional 8% leading to a total average improvement of about 17%.
@article{wick2021onemodel,
abstract = {In this paper, we propose a novel method for Automatic Text Recognition (ATR) on early printed books. Our approach signicantly reduces the Character Error Rates (CERs) for book-specic training when only a few lines of Ground Truth (GT) are available and considerably outperforms previous methods. An ensemble of models is trained simultaneously by optimising each one independently but also with respect
to a fused output obtained by averaging the individual condence matrices. Various experiments on ve early printed books show that this approach already outperforms the current state-of-the-art by up to 20% and 10% on average. Replacing the averaging of the condence matrices during prediction with a condence-based voting boosts our results by an additional 8% leading to a total average improvement of about 17%.},
added-at = {2021-05-02T16:35:57.000+0200},
author = {Wick, Christoph and Reul, Christian},
biburl = {https://www.bibsonomy.org/bibtex/2f969fe96c6af2c5c4ce852357a631e76/chreul},
booktitle = {Proceedings of the 16th International Conference on Document Analysis and Recognition ICDAR 2021},
interhash = {2f76791c05fecd5e3887a12c3c000b42},
intrahash = {f969fe96c6af2c5c4ce852357a631e76},
keywords = {myown myown:selected},
timestamp = {2022-05-28T10:02:06.000+0200},
title = {One-Model Ensemble-Learning for Text Recognition of Historical Printings},
url = {https://link.springer.com/chapter/10.1007/978-3-030-86549-8_25},
year = 2021
}