Publications

• OCR4all 1.0 – Flexible open-source OCR/HTR based on various single-step Solutions. Nöth, Maximilian; Baier, Herbert; Reul, Christian. In 16th IAPR International Workshop On Document Analysis Systems (DAS). 2024.
  • [ Abstract ]
  • [ BibTeX ]
  Recent progress in Automatic Text Recognition (ATR) has been bolstered by the continuous release of open-source implementations addressing individual steps of the workflow. Since the optimal approach varies depending on the material at hand and individual quality requirements, we argue that the next crucial step is the establishment of user-friendly software that facilitate flexible, integrative, and sustainablecombinations of current and future solutions. To this end, we introduce the fully reworked OCR4all which empowers even non-technical users to independently perform ATR achieving high-quality results. OCR4all is 100% open-source and integrates a variety of solutions, ensuring that users can leverage the best tools available for their specific needs.

  @article{noth2024ocr4all, abstract = {Recent progress in Automatic Text Recognition (ATR) has been bolstered by the continuous release of open-source implementations addressing individual steps of the workflow. Since the optimal approach varies depending on the material at hand and individual quality requirements, we argue that the next crucial step is the establishment of user-friendly software that facilitate flexible, integrative, and sustainablecombinations of current and future solutions. To this end, we introduce the fully reworked OCR4all which empowers even non-technical users to independently perform ATR achieving high-quality results. OCR4all is 100% open-source and integrates a variety of solutions, ensuring that users can leverage the best tools available for their specific needs.}, author = {Nöth, Maximilian and Baier, Herbert and Reul, Christian}, journal = {16th IAPR International Workshop On Document Analysis Systems (DAS)}, keywords = {myown:selected}, month = {09}, title = {OCR4all 1.0 – Flexible open-source OCR/HTR based on various single-step Solutions}, year = 2024 }
• Human-Centred Open-Source Automatic Text Recognition for the Humanities with OCR4all. Reul, Christian; Nöth, Maximilian; Baier, Herbert; Chadbourne, Kevin; Langhanki, Florian. In Humanities-Centred AI (CHAI), 4th Workshop at the 47th German Conference on Artificial Intelligence. 2024.
  • [ Abstract ]
  • [ BibTeX ]
  Not least due to the rise of quantitative methods and their need for large volumes of high-quality text, Automatic Text Recognition (ATR) is becoming increasingly important in (Digital) Humanities research. Major technical advances in recent years have been supported by the continuous release of open-source implementations that cover the various individual steps of the ATR workflow. As the optimal approach varies depending on the material at hand and individual quality requirements, the next crucial step is the (further) development of user-friendly software that enables flexible, integrative and sustainable combinations of current and future solutions. To this end, we present the completely revised open-source tool OCR4all, which allows even non-technical users to independently perform an ATR workflow and achieve high-quality results. OCR4all is fully compatible with the ATR solutions from the DFG funding initiative “OCR-D”, so that users have a variety of tools at their disposal that they can use in the best possible way for their specific requirements.

  @article{reul2024humancentred, abstract = {Not least due to the rise of quantitative methods and their need for large volumes of high-quality text, Automatic Text Recognition (ATR) is becoming increasingly important in (Digital) Humanities research. Major technical advances in recent years have been supported by the continuous release of open-source implementations that cover the various individual steps of the ATR workflow. As the optimal approach varies depending on the material at hand and individual quality requirements, the next crucial step is the (further) development of user-friendly software that enables flexible, integrative and sustainable combinations of current and future solutions. To this end, we present the completely revised open-source tool OCR4all, which allows even non-technical users to independently perform an ATR workflow and achieve high-quality results. OCR4all is fully compatible with the ATR solutions from the DFG funding initiative “OCR-D”, so that users have a variety of tools at their disposal that they can use in the best possible way for their specific requirements.}, author = {Reul, Christian and Nöth, Maximilian and Baier, Herbert and Chadbourne, Kevin and Langhanki, Florian}, journal = {Humanities-Centred AI (CHAI), 4th Workshop at the 47th German Conference on Artificial Intelligence}, keywords = {myown}, month = {09}, title = {Human-Centred Open-Source Automatic Text Recognition for the Humanities with OCR4all}, year = 2024 }
• OCR(-D)4all – An easy to use and highly adaptable Open-Source Solution for Automatic Text Recognition of Historical Printings and Manuscripts. Nöth, Maximilian; Baier, Herbert; Reul, Christian. In 4th Conference for Research Software Engineering in Germany (deRSE). 2024.
  • [ BibTeX ]
  • [ URL ]
  @article{noth2024ocrd4all, author = {Nöth, Maximilian and Baier, Herbert and Reul, Christian}, journal = {4th Conference for Research Software Engineering in Germany (deRSE)}, keywords = {myown}, month = {04}, title = {OCR(-D)4all – An easy to use and highly adaptable Open-Source Solution for Automatic Text Recognition of Historical Printings and Manuscripts}, year = 2024 }
• Edierst Du noch oder trainierst Du schon? Forschungsdaten als Grundlage von Trainingsdaten für die automatische Texterkennung. Boenig, Matthias; Baierer, Konstantin; Hinrichsen, Lena; Würzner, Kay-Michael; Reul, Christian. In DHd2024: Quo Vadis DH. 2024.
  • [ BibTeX ]
  • [ URL ]
  @article{boenig2024edierst, author = {Boenig, Matthias and Baierer, Konstantin and Hinrichsen, Lena and Würzner, Kay-Michael and Reul, Christian}, journal = {DHd2024: Quo Vadis DH}, keywords = {myown}, month = {03}, title = {Edierst Du noch oder trainierst Du schon? Forschungsdaten als Grundlage von Trainingsdaten für die automatische Texterkennung}, year = 2024 }
• Das richtige Tool für die Volltextdigitalisierung. Baierer, Konstantin; Hinrichsen, Lena; Boenig, Matthias; Reul, Christian; Sautter, Lilja; Mustafa, Mehmed; Will, Larissa. In DHd2024: Quo Vadis DH. 2024.
  • [ BibTeX ]
  • [ URL ]
  @article{baierer2024richtige, author = {Baierer, Konstantin and Hinrichsen, Lena and Boenig, Matthias and Reul, Christian and Sautter, Lilja and Mustafa, Mehmed and Will, Larissa}, journal = {DHd2024: Quo Vadis DH}, keywords = {myown}, title = {Das richtige Tool für die Volltextdigitalisierung}, year = 2024 }

• Analyse, Produktion, Reflexion: Nachnutzungsszenarien für Forschungsdaten am Beispiel der Daten des Projekts Dehmel digital. Bläß, Sandra; Flüh, Marie; Nantke, Julia; Reul, Christian. In DHd2023: Open Humanities, Open Culture. 2023.
  • [ BibTeX ]
  • [ URL ]
  @article{blass2023analyse, author = {Bläß, Sandra and Flüh, Marie and Nantke, Julia and Reul, Christian}, journal = {DHd2023: Open Humanities, Open Culture}, keywords = {myown}, title = {Analyse, Produktion, Reflexion: Nachnutzungsszenarien für Forschungsdaten am Beispiel der Daten des Projekts Dehmel digital}, year = 2023 }
• Algorithmen-gestützte Analyse visuell-materieller Eigenschaften von Briefen. Nantke, Julia; Reul, Christian; Bläß, Sandra; Flüh, Marie. In DHd2023: Open Humanities, Open Culture. 2023.
  • [ BibTeX ]
  • [ URL ]
  @article{nantke2023algorithmengesttzte, author = {Nantke, Julia and Reul, Christian and Bläß, Sandra and Flüh, Marie}, journal = {DHd2023: Open Humanities, Open Culture}, keywords = {myown}, title = {Algorithmen-gestützte Analyse visuell-materieller Eigenschaften von Briefen}, year = 2023 }
• Synoptische Interfaces Digitaler Editionen. Herbst, Yannik; Roeder, Torsten; Reul, Christian. In DHd2023: Open Humanities, Open Culture. 2023.
  • [ BibTeX ]
  • [ URL ]
  @article{herbst2023synoptische, author = {Herbst, Yannik and Roeder, Torsten and Reul, Christian}, journal = {DHd2023: Open Humanities, Open Culture}, keywords = {myown}, title = {Synoptische Interfaces Digitaler Editionen}, year = 2023 }

• Handwritten Text Recognition und Word Mover’s Distance als Grundlagen der digitalen Edition "Die Kindheit Jesu Konrads von Fußesbrunnen". Tomasek, Stefan; Reul, Christian; Wehner, Maximilian. In DHd2022: Kulturen des digitalen Gedächtnisses. 2022.
  • [ BibTeX ]
  • [ URL ]
  @article{tomasek2022handwritten, author = {Tomasek, Stefan and Reul, Christian and Wehner, Maximilian}, journal = {DHd2022: Kulturen des digitalen Gedächtnisses}, keywords = {myown}, title = {Handwritten Text Recognition und Word Mover’s Distance als Grundlagen der digitalen Edition "Die Kindheit Jesu Konrads von Fußesbrunnen"}, year = 2022 }
• Semantische Tiefenerschließung historischer Lexika mittels Text- und Typographieerkennung. Reul, Christian. In Ursula Rautenberg/Anja Voeste (Hgg.): Typographie. Theoretische Konzeptionen, historische Perspektiven, künstlerische Applikationen. 2022.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  Bei der Konvertierung typographisch komplexer Textvorlagen in eine elektronische Form stellt die rein textuelle Erfassung, z. B. mittels Optical Character Recognition (OCR), häufig nur den ersten Bearbeitungsschritt dar. Um die logische Struktur solcher Texte zu erschließen, ist zudem eine präzise automatische Erkennung der enthaltenen typographischen Attribute nötig. Dieser Artikel stellt eine Methode vor, die durch Training einer frei verfügbaren Texterkennungssoftware eine feinkörnige Klassifikation typographischer Auszeichnungsmittel ermöglicht und die so gewonnenen semantischen Informationen auf das Ergebnis der textuellen Erkennung abbildet. Als Anwendungsbeispiel dient Daniel Sanders‘ Wörterbuch der Deutschen Sprache, in dem der Typographie eine besonders wichtige und komplexe semantische Funktion zukommt.

  @article{reul2021semantische, abstract = {Bei der Konvertierung typographisch komplexer Textvorlagen in eine elektronische Form stellt die rein textuelle Erfassung, z. B. mittels Optical Character Recognition (OCR), häufig nur den ersten Bearbeitungsschritt dar. Um die logische Struktur solcher Texte zu erschließen, ist zudem eine präzise automatische Erkennung der enthaltenen typographischen Attribute nötig. Dieser Artikel stellt eine Methode vor, die durch Training einer frei verfügbaren Texterkennungssoftware eine feinkörnige Klassifikation typographischer Auszeichnungsmittel ermöglicht und die so gewonnenen semantischen Informationen auf das Ergebnis der textuellen Erkennung abbildet. Als Anwendungsbeispiel dient Daniel Sanders‘ Wörterbuch der Deutschen Sprache, in dem der Typographie eine besonders wichtige und komplexe semantische Funktion zukommt.}, author = {Reul, Christian}, journal = {Ursula Rautenberg/Anja Voeste (Hgg.): Typographie. Theoretische Konzeptionen, historische Perspektiven, künstlerische Applikationen}, keywords = {myown}, title = {Semantische Tiefenerschließung historischer Lexika mittels Text- und Typographieerkennung}, year = 2022 }
• Open Source Handwritten Text Recognition on Medieval Manuscripts using Mixed Models and Document-Specific Finetuning. Reul, Christian; Tomasek, Stefan; Langhanki, Florian; Springmann, Uwe. In 15th IAPR International Workshop on Document Analysis Systems (DAS). 2022.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  This paper deals with the task of practical and open source Handwritten Text Recognition (HTR) on German medieval manuscripts. We report on our efforts to construct mixed recognition models which can be applied out-of-the-box without any further document-specific training but also serve as a starting point for finetuning by training a new model on a few pages of transcribed text (ground truth). To train the mixed models we collected a corpus of 35 manuscripts and ca. 12.5k text lines for two widely used handwriting styles, Gothic and Bastarda cursives. Evaluating the mixed models out-of-the-box on four unseen manuscripts resulted in an average Character Error Rate (CER) of 6.22%. After training on 2, 4 and eventually 32 pages the CER dropped to 3.27%, 2.58%, and 1.65%, respectively. While the in-domain recognition and training of models (Bastarda model to Bastarda material, Gothic to Gothic) unsurprisingly yielded the best results, finetuning out-of-domain models to unseen scripts was still shown to be superior to training from scratch. Our new mixed models have been made openly available to the community.

  @article{reul2022source, abstract = {This paper deals with the task of practical and open source Handwritten Text Recognition (HTR) on German medieval manuscripts. We report on our efforts to construct mixed recognition models which can be applied out-of-the-box without any further document-specific training but also serve as a starting point for finetuning by training a new model on a few pages of transcribed text (ground truth). To train the mixed models we collected a corpus of 35 manuscripts and ca. 12.5k text lines for two widely used handwriting styles, Gothic and Bastarda cursives. Evaluating the mixed models out-of-the-box on four unseen manuscripts resulted in an average Character Error Rate (CER) of 6.22%. After training on 2, 4 and eventually 32 pages the CER dropped to 3.27%, 2.58%, and 1.65%, respectively. While the in-domain recognition and training of models (Bastarda model to Bastarda material, Gothic to Gothic) unsurprisingly yielded the best results, finetuning out-of-domain models to unseen scripts was still shown to be superior to training from scratch. Our new mixed models have been made openly available to the community.}, author = {Reul, Christian and Tomasek, Stefan and Langhanki, Florian and Springmann, Uwe}, journal = {15th IAPR International Workshop on Document Analysis Systems (DAS)}, keywords = {myown:selected}, title = {Open Source Handwritten Text Recognition on Medieval Manuscripts using Mixed Models and Document-Specific Finetuning}, year = 2022 }
• OCR4all - Massenvolltextdigitalisierung von Drucken mithilfe von OCR-D und hochqualitative Transkription von Handschriften. Langhanki, Florian; Wehner, Maximilian; Baierer, Konstantin; Hinrichsen, Lena; Reul, Christian. In DHd2022: Kulturen des digitalen Gedächtnisses. 2022.
  • [ BibTeX ]
  • [ URL ]
  @article{langhanki2022ocr4all, author = {Langhanki, Florian and Wehner, Maximilian and Baierer, Konstantin and Hinrichsen, Lena and Reul, Christian}, journal = {DHd2022: Kulturen des digitalen Gedächtnisses}, keywords = {myown}, title = {OCR4all - Massenvolltextdigitalisierung von Drucken mithilfe von OCR-D und hochqualitative Transkription von Handschriften}, year = 2022 }

• One-Model Ensemble-Learning for Text Recognition of Historical Printings. Wick, Christoph; Reul, Christian. In 16th International Conference on Document Analysis and Recognition (ICDAR). 2021.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  In this paper, we propose a novel method for Automatic Text Recognition (ATR) on early printed books. Our approach signi�cantly reduces the Character Error Rates (CERs) for book-speci�c 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 con�dence 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 con�dence matrices during prediction with a con�dence-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 signi�cantly reduces the Character Error Rates (CERs) for book-speci�c 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 con�dence 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 con�dence matrices during prediction with a con�dence-based voting boosts our results by an additional 8% leading to a total average improvement of about 17%.}, author = {Wick, Christoph and Reul, Christian}, booktitle = {16th International Conference on Document Analysis and Recognition (ICDAR)}, keywords = {myown:selected}, title = {One-Model Ensemble-Learning for Text Recognition of Historical Printings}, year = 2021 }
• OCR-D & OCR4all: Two Complementary Approaches for Improved OCR of Historical Sources. Baierer, Konstantin; Büttner, Andreas; Engl, Elisabeth; Hinrichsen, Lena; Reul, Christian. In 6th International Workshop on Computational History. 2021.
  • [ Abstract ]
  • [ BibTeX ]
  In order to leverage recent advances within the field of computer science for tasks ranging from full text search to various forms of quantitative analysis, an important desideratum in many areas of historical research is the availability of sources in the form of machine-actionable text. Here we present two complementary and free-to-use approaches, OCR-D and OCR4all. They deal with the conversion of scanned historical documents into such machine-actionable text and allow to tackle a variety of different use cases with diverse user requirements regarding both quantity and quality.

  @article{baierer2021ocr4all, abstract = {In order to leverage recent advances within the field of computer science for tasks ranging from full text search to various forms of quantitative analysis, an important desideratum in many areas of historical research is the availability of sources in the form of machine-actionable text. Here we present two complementary and free-to-use approaches, OCR-D and OCR4all. They deal with the conversion of scanned historical documents into such machine-actionable text and allow to tackle a variety of different use cases with diverse user requirements regarding both quantity and quality.}, author = {Baierer, Konstantin and Büttner, Andreas and Engl, Elisabeth and Hinrichsen, Lena and Reul, Christian}, journal = {6th International Workshop on Computational History}, keywords = {myown}, title = {OCR-D & OCR4all: Two Complementary Approaches for Improved OCR of Historical Sources}, year = 2021 }
• Mixed Model OCR Training on Historical Latin Script for Out-of-the-Box Recognition and Finetuning. Reul, Christian; Wick, Christoph; Nöth, Maximilian; Büttner, Andreas; Wehner, Maximilian; Springmann, Uwe. In 6th International Workshop on Historical Document Imaging and Processing (HIP). 2021.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  In order to apply Optical Character Recognition (OCR) to historical printings of Latin script fully automatically, we report on our efforts to construct a widely-applicable polyfont recognition model yielding text with a Character Error Rate (CER) around 2% when applied out-of-the-box. Moreover, we show how this model can be further finetuned to specific classes of printings with little manual and computational effort. The mixed or polyfont model is trained on a wide variety of materials, in terms of age (from the 15th to the 19th century), typography (various types of Fraktur and Antiqua), and languages (among others, German, Latin, and French). To optimize the results we combined established techniques of OCR training like pretraining, data augmentation, and voting. In addition, we used various preprocessing methods to enrich the training data and obtain more robust models. We also implemented a two-stage approach which first trains on all available, considerably unbalanced data and then refines the output by training on a selected more balanced subset. Evaluations on 29 previously unseen books resulted in a CER of 1.73%, outperforming a widely used standard model with a CER of 2.84% by almost 40%. Training a more specialized model for some unseen Early Modern Latin books starting from our mixed model led to a CER of 1.47%, an improvement of up to 50% compared to training from scratch and up to 30% compared to training from the aforementioned standard model. Our new mixed model is made openly available to the community.

  @article{reul2021mixed, abstract = {In order to apply Optical Character Recognition (OCR) to historical printings of Latin script fully automatically, we report on our efforts to construct a widely-applicable polyfont recognition model yielding text with a Character Error Rate (CER) around 2% when applied out-of-the-box. Moreover, we show how this model can be further finetuned to specific classes of printings with little manual and computational effort. The mixed or polyfont model is trained on a wide variety of materials, in terms of age (from the 15th to the 19th century), typography (various types of Fraktur and Antiqua), and languages (among others, German, Latin, and French). To optimize the results we combined established techniques of OCR training like pretraining, data augmentation, and voting. In addition, we used various preprocessing methods to enrich the training data and obtain more robust models. We also implemented a two-stage approach which first trains on all available, considerably unbalanced data and then refines the output by training on a selected more balanced subset. Evaluations on 29 previously unseen books resulted in a CER of 1.73%, outperforming a widely used standard model with a CER of 2.84% by almost 40%. Training a more specialized model for some unseen Early Modern Latin books starting from our mixed model led to a CER of 1.47%, an improvement of up to 50% compared to training from scratch and up to 30% compared to training from the aforementioned standard model. Our new mixed model is made openly available to the community.}, author = {Reul, Christian and Wick, Christoph and Nöth, Maximilian and Büttner, Andreas and Wehner, Maximilian and Springmann, Uwe}, journal = {6th International Workshop on Historical Document Imaging and Processing (HIP)}, keywords = {myown:selected}, title = {Mixed Model OCR Training on Historical Latin Script for Out-of-the-Box Recognition and Finetuning}, year = 2021 }

• Calamari - A High-Performance Tensorflow-based Deep Learning Package for Optical Character Recognition. Wick, Christoph; Reul, Christian; Puppe, Frank. In Digital Humanities Quarterly, 14(2). 2020.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  Optical Character Recognition (OCR) on contemporary and historical data is still in the focus of many researchers. Especially historical prints require book specific trained OCR models to achieve applicable results [Springmann and Lüdeling 2017] [Reul et al. 2017a]. To reduce the human effort for manually annotating ground truth (GT) various techniques such as voting and pretraining have shown to be very efficient [Reul et al. 2018a] [Reul et al. 2018b]. Calamari is a new open source OCR line recognition software that both uses state-of-the art Deep Neural Networks (DNNs) implemented in Tensorflow and giving native support for techniques such as pretraining and voting. The customizable network architectures constructed of Convolutional Neural Networks (CNNS) and Long-Short-Term-Memory (LSTM) layers are trained by the so-called Connectionist Temporal Classification (CTC) algorithm of Graves et al. (2006). Optional usage of a GPU drastically reduces the computation times for both training and prediction. We use two different datasets to compare the performance of Calamari to OCRopy, OCRopus3, and Tesseract 4. Calamari reaches a Character Error Rate (CER) of 0.11% on the UW3 dataset written in modern English and 0.18% on the DTA19 dataset written in German Fraktur, which considerably outperforms the results of the existing softwares.

  @article{wick2020calamari, abstract = {Optical Character Recognition (OCR) on contemporary and historical data is still in the focus of many researchers. Especially historical prints require book specific trained OCR models to achieve applicable results [Springmann and Lüdeling 2017] [Reul et al. 2017a]. To reduce the human effort for manually annotating ground truth (GT) various techniques such as voting and pretraining have shown to be very efficient [Reul et al. 2018a] [Reul et al. 2018b]. Calamari is a new open source OCR line recognition software that both uses state-of-the art Deep Neural Networks (DNNs) implemented in Tensorflow and giving native support for techniques such as pretraining and voting. The customizable network architectures constructed of Convolutional Neural Networks (CNNS) and Long-Short-Term-Memory (LSTM) layers are trained by the so-called Connectionist Temporal Classification (CTC) algorithm of Graves et al. (2006). Optional usage of a GPU drastically reduces the computation times for both training and prediction. We use two different datasets to compare the performance of Calamari to OCRopy, OCRopus3, and Tesseract 4. Calamari reaches a Character Error Rate (CER) of 0.11% on the UW3 dataset written in modern English and 0.18% on the DTA19 dataset written in German Fraktur, which considerably outperforms the results of the existing softwares.}, author = {Wick, Christoph and Reul, Christian and Puppe, Frank}, journal = {Digital Humanities Quarterly}, keywords = {sys:relevantfor:uniwue_info6}, number = 2, title = {Calamari - A High-Performance Tensorflow-based Deep Learning Package for Optical Character Recognition}, volume = 14, year = 2020 }

• State of the Art Optical Character Recognition of 19th Century Fraktur Scripts using Open Source Engines. Reul, Christian; Springmann, Uwe; Wick, Christoph; Puppe, Frank. In DHd 2019 Digital Humanities: multimedial & multimodal. 2019.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  In this paper we evaluate Optical Character Recognition (OCR) of 19th century Fraktur scripts without book-specific training using mixed models, i.e. models trained to recognize a variety of fonts and typesets from previously unseen sources. We describe the training process leading to strong mixed OCR models and compare them to freely available models of the popular open source engines OCRopus and Tesseract as well as the commercial state of the art system ABBYY. For evaluation, we use a varied collection of unseen data from books, journals, and a dictionary from the 19th century. The experiments show that training mixed models with real data is superior to training with synthetic data and that the novel OCR engine Calamari outperforms the other engines considerably, on average reducing ABBYYs character error rate (CER) by over 70%, resulting in an average CER below 1%.

  @article{reul2019state, abstract = {In this paper we evaluate Optical Character Recognition (OCR) of 19th century Fraktur scripts without book-specific training using mixed models, i.e. models trained to recognize a variety of fonts and typesets from previously unseen sources. We describe the training process leading to strong mixed OCR models and compare them to freely available models of the popular open source engines OCRopus and Tesseract as well as the commercial state of the art system ABBYY. For evaluation, we use a varied collection of unseen data from books, journals, and a dictionary from the 19th century. The experiments show that training mixed models with real data is superior to training with synthetic data and that the novel OCR engine Calamari outperforms the other engines considerably, on average reducing ABBYYs character error rate (CER) by over 70%, resulting in an average CER below 1%.}, author = {Reul, Christian and Springmann, Uwe and Wick, Christoph and Puppe, Frank}, journal = {DHd 2019 Digital Humanities: multimedial & multimodal}, keywords = {sys:relevantfor:uniwue_info6}, title = {State of the Art Optical Character Recognition of 19th Century Fraktur Scripts using Open Source Engines}, year = 2019 }
• Ground Truth for training OCR engines on historical documents in German Fraktur and Early Modern Latin. Springmann, Uwe; Reul, Christian; Dipper, Stefanie; Baiter, Johannes. In JLCL: Special Issue on Automatic Text and Layout Recognition. 2019.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  In this paper we describe a dataset of German and Latin ground truth (GT) for historical OCR in the form of printed text line images paired with their transcription. This dataset, called GT4HistOCR, consists of 313,173 line pairs covering a wide period of printing dates from incunabula from the 15th century to 19th century books printed in Fraktur types and is openly available under a CC-BY ⒋0 license. The special form of GT as line image/transcription pairs makes it directly usable to train state-of-the-art recognition models for OCR software employing recurring neural networks in LSTM architecture such as Tesseract 4 or OCRopus. We also provide some pretrained OCRopus models for subcorpora of our dataset yielding between 95% (early printings) and 98% (19th century Fraktur printings) character accuracy rates on unseen test cases, a Perl script to harmonize GT produced by different transcription rules, and give hints on how to construct GT for OCR purposes which has requirements that may differ from linguistically motivated transcriptions.

  @article{springmann2019ground, abstract = {In this paper we describe a dataset of German and Latin ground truth (GT) for historical OCR in the form of printed text line images paired with their transcription. This dataset, called GT4HistOCR, consists of 313,173 line pairs covering a wide period of printing dates from incunabula from the 15th century to 19th century books printed in Fraktur types and is openly available under a CC-BY ⒋0 license. The special form of GT as line image/transcription pairs makes it directly usable to train state-of-the-art recognition models for OCR software employing recurring neural networks in LSTM architecture such as Tesseract 4 or OCRopus. We also provide some pretrained OCRopus models for subcorpora of our dataset yielding between 95% (early printings) and 98% (19th century Fraktur printings) character accuracy rates on unseen test cases, a Perl script to harmonize GT produced by different transcription rules, and give hints on how to construct GT for OCR purposes which has requirements that may differ from linguistically motivated transcriptions.}, author = {Springmann, Uwe and Reul, Christian and Dipper, Stefanie and Baiter, Johannes}, journal = {JLCL: Special Issue on Automatic Text and Layout Recognition}, keywords = {sys:relevantfor:uniwue_info6}, title = {Ground Truth for training OCR engines on historical documents in German Fraktur and Early Modern Latin}, year = 2019 }
• OCR4all - An Open-Source Tool Providing a (Semi-)Automatic OCR Workflow for Historical Printings. Reul, Christian; Christ, Dennis; Hartelt, Alexander; Balbach, Nico; Wehner, Maximilian; Springmann, Uwe; Wick, Christoph; Grundig, Christine; Büttner, Andreas; Puppe, Frank. In Applied Sciences, 9(22). 2019.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  Optical Character Recognition (OCR) on historical printings is a challenging task mainly due to the complexity of the layout and the highly variant typography. Nevertheless, in the last few years, great progress has been made in the area of historical OCR, resulting in several powerful open-source tools for preprocessing, layout analysis and segmentation, character recognition, and post-processing. The drawback of these tools often is their limited applicability by non-technical users like humanist scholars and in particular the combined use of several tools in a workflow. In this paper, we present an open-source OCR software called OCR4all, which combines state-of-the-art OCR components and continuous model training into a comprehensive workflow. While a variety of materials can already be processed fully automatically, books with more complex layouts require manual intervention by the users. This is mostly due to the fact that the required ground truth for training stronger mixed models (for segmentation, as well as text recognition) is not available, yet, neither in the desired quantity nor quality. To deal with this issue in the short run, OCR4all offers a comfortable GUI that allows error corrections not only in the final output, but already in early stages to minimize error propagations. In the long run, this constant manual correction produces large quantities of valuable, high quality training material, which can be used to improve fully automatic approaches. Further on, extensive configuration capabilities are provided to set the degree of automation of the workflow and to make adaptations to the carefully selected default parameters for specific printings, if necessary. During experiments, the fully automated application on 19th Century novels showed that OCR4all can considerably outperform the commercial state-of-the-art tool ABBYY Finereader on moderate layouts if suitably pretrained mixed OCR models are available. Furthermore, on very complex early printed books, even users with minimal or no experience were able to capture the text with manageable effort and great quality, achieving excellent Character Error Rates (CERs) below 0.5%. The architecture of OCR4all allows the easy integration (or substitution) of newly developed tools for its main components by standardized interfaces like PageXML, thus aiming at continual higher automation for historical printings.

  @article{reul2019ocr4all, abstract = {Optical Character Recognition (OCR) on historical printings is a challenging task mainly due to the complexity of the layout and the highly variant typography. Nevertheless, in the last few years, great progress has been made in the area of historical OCR, resulting in several powerful open-source tools for preprocessing, layout analysis and segmentation, character recognition, and post-processing. The drawback of these tools often is their limited applicability by non-technical users like humanist scholars and in particular the combined use of several tools in a workflow. In this paper, we present an open-source OCR software called OCR4all, which combines state-of-the-art OCR components and continuous model training into a comprehensive workflow. While a variety of materials can already be processed fully automatically, books with more complex layouts require manual intervention by the users. This is mostly due to the fact that the required ground truth for training stronger mixed models (for segmentation, as well as text recognition) is not available, yet, neither in the desired quantity nor quality. To deal with this issue in the short run, OCR4all offers a comfortable GUI that allows error corrections not only in the final output, but already in early stages to minimize error propagations. In the long run, this constant manual correction produces large quantities of valuable, high quality training material, which can be used to improve fully automatic approaches. Further on, extensive configuration capabilities are provided to set the degree of automation of the workflow and to make adaptations to the carefully selected default parameters for specific printings, if necessary. During experiments, the fully automated application on 19th Century novels showed that OCR4all can considerably outperform the commercial state-of-the-art tool ABBYY Finereader on moderate layouts if suitably pretrained mixed OCR models are available. Furthermore, on very complex early printed books, even users with minimal or no experience were able to capture the text with manageable effort and great quality, achieving excellent Character Error Rates (CERs) below 0.5%. The architecture of OCR4all allows the easy integration (or substitution) of newly developed tools for its main components by standardized interfaces like PageXML, thus aiming at continual higher automation for historical printings.}, author = {Reul, Christian and Christ, Dennis and Hartelt, Alexander and Balbach, Nico and Wehner, Maximilian and Springmann, Uwe and Wick, Christoph and Grundig, Christine and Büttner, Andreas and Puppe, Frank}, journal = {Applied Sciences}, keywords = {sys:relevantfor:uniwue_info6}, number = 22, title = {OCR4all - An Open-Source Tool Providing a (Semi-)Automatic OCR Workflow for Historical Printings}, volume = 9, year = 2019 }
• Korrektur von fehlerhaften OCR Ergebnissen durch automatisches Alignment mit Texten eines Korpus. Bald, Markus; Damiani, Vincenzo; Essler, Holger; Eyeselein, Björn; Reul, Christian; Puppe, Frank. In DHd2019: multimedial & multimodal. 2019.
  • [ BibTeX ]
  • [ URL ]
  @article{bald2019korrektur, author = {Bald, Markus and Damiani, Vincenzo and Essler, Holger and Eyeselein, Björn and Reul, Christian and Puppe, Frank}, journal = {DHd2019: multimedial & multimodal}, keywords = {myown}, title = {Korrektur von fehlerhaften OCR Ergebnissen durch automatisches Alignment mit Texten eines Korpus}, year = 2019 }
• Automatic Semantic Text Tagging on Historical Lexica by Combining OCR and Typography Classification. Reul, Christian; Göttel, Sebastian; Springmann, Uwe; Wick, Christoph; Würzner, Kay-Michael; Puppe, Frank. In 3rd International Conference on Digital Access to Textual Cultural Heritage (DATeCH). 2019.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  When converting historical lexica into electronic form the goal is not only to obtain a high quality OCR result for the text but also to perform a precise automatic recognition of typographical attributes in order to capture the logical structure. For that purpose, we present a method that enables a fine-grained typography classification by training an open source OCR engine both on traditional OCR and typography recognition and show how to map the obtained typography information to the OCR recognized text output. As a test case, we used a German dictionary (Sander's "Wörterbuch der Deutschen Sprache") from the 19th century, which comprises a particularly complex semantic function of typography. Despite the very challenging material, we achieved a character error rate below 0.4% and a typography recognition that assigns the correct label to close to 99% of the words. In contrast to many existing methods, our novel approach works with real historical data and can deal with frequent typography changes even within lines.

  @article{reul2019automatic, abstract = {When converting historical lexica into electronic form the goal is not only to obtain a high quality OCR result for the text but also to perform a precise automatic recognition of typographical attributes in order to capture the logical structure. For that purpose, we present a method that enables a fine-grained typography classification by training an open source OCR engine both on traditional OCR and typography recognition and show how to map the obtained typography information to the OCR recognized text output. As a test case, we used a German dictionary (Sander's "Wörterbuch der Deutschen Sprache") from the 19th century, which comprises a particularly complex semantic function of typography. Despite the very challenging material, we achieved a character error rate below 0.4% and a typography recognition that assigns the correct label to close to 99% of the words. In contrast to many existing methods, our novel approach works with real historical data and can deal with frequent typography changes even within lines.}, author = {Reul, Christian and Göttel, Sebastian and Springmann, Uwe and Wick, Christoph and Würzner, Kay-Michael and Puppe, Frank}, journal = {3rd International Conference on Digital Access to Textual Cultural Heritage (DATeCH)}, keywords = {myown:selected}, title = {Automatic Semantic Text Tagging on Historical Lexica by Combining OCR and Typography Classification}, year = 2019 }

• Comparison of OCR Accuracy on Early Printed Books using the Open Source Engines Calamari and OCRopus. Wick, Christoph; Reul, Christian; Puppe, Frank. In JLCL: Special Issue on Automatic Text and Layout Recognition, 33(1), pp. 79–96. 2018.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  This paper proposes a combination of a convolutional and an LSTM network to improve the accuracy of OCR on early printed books. While the default approach of line based OCR is to use a single LSTM layer as provided by the well-established OCR software OCRopus (OCRopy), we utilize a CNN- and Pooling-Layer combination in advance of an LSTM layer as implemented by the novel OCR software Calamari. Since historical prints often require book specific models trained on manually labeled ground truth (GT) the goal is to maximize the recognition accuracy of a trained model while keeping the needed manual effort to a minimum. We show, that the deep model significantly outperforms the shallow LSTM network when using both many and only a few training examples, although the deep network has a higher amount of trainable parameters. Hereby, the error rate is reduced by a factor of up to 55%, yielding character error rates (CER) of 1% and below for 1,000 lines of training. To further improve the results, we apply a confidence voting mechanism to achieve CERs below 0.5%. A simple data augmentation scheme and the usage of pretrained models reduces the CER further by up to 62% if only few training data is available. Thus, we require only 100 lines of GT to reach an average CER of 1.2%. The runtime of the deep model for training and prediction of a book behaves very similar to a shallow network when trained on a CPU. However, the usage of a GPU, as supported by Calamari, reduces the prediction time by a factor of at least four and the training time by more than six.

  @article{wick2018comparison, abstract = {This paper proposes a combination of a convolutional and an LSTM network to improve the accuracy of OCR on early printed books. While the default approach of line based OCR is to use a single LSTM layer as provided by the well-established OCR software OCRopus (OCRopy), we utilize a CNN- and Pooling-Layer combination in advance of an LSTM layer as implemented by the novel OCR software Calamari. Since historical prints often require book specific models trained on manually labeled ground truth (GT) the goal is to maximize the recognition accuracy of a trained model while keeping the needed manual effort to a minimum. We show, that the deep model significantly outperforms the shallow LSTM network when using both many and only a few training examples, although the deep network has a higher amount of trainable parameters. Hereby, the error rate is reduced by a factor of up to 55%, yielding character error rates (CER) of 1% and below for 1,000 lines of training. To further improve the results, we apply a confidence voting mechanism to achieve CERs below 0.5%. A simple data augmentation scheme and the usage of pretrained models reduces the CER further by up to 62% if only few training data is available. Thus, we require only 100 lines of GT to reach an average CER of 1.2%. The runtime of the deep model for training and prediction of a book behaves very similar to a shallow network when trained on a CPU. However, the usage of a GPU, as supported by Calamari, reduces the prediction time by a factor of at least four and the training time by more than six.}, author = {Wick, Christoph and Reul, Christian and Puppe, Frank}, journal = {JLCL: Special Issue on Automatic Text and Layout Recognition}, keywords = {sys:relevantfor:uniwue_info6}, number = 1, pages = {79-96}, title = {Comparison of OCR Accuracy on Early Printed Books using the Open Source Engines Calamari and OCRopus}, volume = 33, year = 2018 }
• Improving OCR Accuracy on Early Printed Books by combining Pretraining, Voting, and Active Learning. Reul, Christian; Springmann, Uwe; Wick, Christoph; Puppe, Frank. In JLCL: Special Issue on Automatic Text and Layout Recognition, 33(1), pp. 3–24. 2018.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  We combine three methods which significantly improve the OCR accuracy of OCR models trained on early printed books: (1) The pretraining method utilizes the information stored in already existing models trained on a variety of typesets (mixed models) instead of starting the training from scratch. (2) Performing cross fold training on a single set of ground truth data (line images and their transcriptions) with a single OCR engine (OCRopus) produces a committee whose members then vote for the best outcome by also taking the top-N alternatives and their intrinsic confidence values into account. (3) Following the principle of maximal disagreement we select additional training lines which the voters disagree most on, expecting them to offer the highest information gain for a subsequent training (active learning). Evaluations on six early printed books yielded the following results: On average the combination of pretraining and voting improved the character accuracy by 46% when training five folds starting from the same mixed model. This number rose to 53% when using different models for pretraining, underlining the importance of diverse voters. Incorporating active learning improved the obtained results by another 16% on average (evaluated on three of the six books). Overall, the proposed methods lead to an average error rate of 2.5% when training on only 60 lines. Using a substantial ground truth pool of 1,000 lines brought the error rate down even further to less than 1% on average.

  @article{reul2018improving, abstract = {We combine three methods which significantly improve the OCR accuracy of OCR models trained on early printed books: (1) The pretraining method utilizes the information stored in already existing models trained on a variety of typesets (mixed models) instead of starting the training from scratch. (2) Performing cross fold training on a single set of ground truth data (line images and their transcriptions) with a single OCR engine (OCRopus) produces a committee whose members then vote for the best outcome by also taking the top-N alternatives and their intrinsic confidence values into account. (3) Following the principle of maximal disagreement we select additional training lines which the voters disagree most on, expecting them to offer the highest information gain for a subsequent training (active learning). Evaluations on six early printed books yielded the following results: On average the combination of pretraining and voting improved the character accuracy by 46% when training five folds starting from the same mixed model. This number rose to 53% when using different models for pretraining, underlining the importance of diverse voters. Incorporating active learning improved the obtained results by another 16% on average (evaluated on three of the six books). Overall, the proposed methods lead to an average error rate of 2.5% when training on only 60 lines. Using a substantial ground truth pool of 1,000 lines brought the error rate down even further to less than 1% on average.}, author = {Reul, Christian and Springmann, Uwe and Wick, Christoph and Puppe, Frank}, journal = {JLCL: Special Issue on Automatic Text and Layout Recognition}, keywords = {sys:relevantfor:uniwue_info6}, number = 1, pages = {3-24}, title = {Improving OCR Accuracy on Early Printed Books by combining Pretraining, Voting, and Active Learning}, volume = 33, year = 2018 }
• Improving OCR Accuracy on Early Printed Books by Utilizing Cross Fold Training and Voting. Reul, Christian; Springmann, Uwe; Wick, Christoph; Puppe, Frank. In 13th IAPR International Workshop on Document Analysis Systems (DAS). 2018.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  In this paper we introduce a method that significantly reduces the character error rates for OCR text obtained from OCRopus models trained on early printed books. The method uses a combination of cross fold training and confidence based voting. After allocating the available ground truth in different subsets several training processes are performed, each resulting in a specific OCR model. The OCR text generated by these models then gets voted to determine the final output by taking the recognized characters, their alternatives, and the confidence values assigned to each character into consideration. Experiments on seven early printed books show that the proposed method outperforms the standard approach considerably by reducing the amount of errors by up to 50% and more.

  @article{reul2018voting, abstract = {In this paper we introduce a method that significantly reduces the character error rates for OCR text obtained from OCRopus models trained on early printed books. The method uses a combination of cross fold training and confidence based voting. After allocating the available ground truth in different subsets several training processes are performed, each resulting in a specific OCR model. The OCR text generated by these models then gets voted to determine the final output by taking the recognized characters, their alternatives, and the confidence values assigned to each character into consideration. Experiments on seven early printed books show that the proposed method outperforms the standard approach considerably by reducing the amount of errors by up to 50% and more.}, author = {Reul, Christian and Springmann, Uwe and Wick, Christoph and Puppe, Frank}, journal = {13th IAPR International Workshop on Document Analysis Systems (DAS)}, keywords = {myown:selected}, title = {Improving OCR Accuracy on Early Printed Books by Utilizing Cross Fold Training and Voting}, year = 2018 }

• Transfer Learning for OCRopus Model Training on Early Printed Books. Reul, Christian; Wick, Christoph; Springmann, Uwe; Puppe, Frank. In 027.7 Journal for Library Culture. 2017.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  A method is presented that significantly reduces the character error rates for OCR text obtained from OCRopus models trained on early printed books when only small amounts of diplomatic transcriptions are available. This is achieved by building from already existing models during training instead of starting from scratch. To overcome the discrepancies between the set of characters of the pretrained model and the additional ground truth the OCRopus code is adapted to allow for alphabet expansion or reduction. The character set is now capable of flexibly adding and deleting characters from the pretrained alphabet when an existing model is loaded. For our experiments we use a self-trained mixed model on early Latin prints and the two standard OCRopus models on modern English and German Fraktur texts. The evaluation on seven early printed books showed that training from the Latin mixed model reduces the average amount of errors by 43% and 26%, compared to training from scratch with 60 and 150 lines of ground truth, respectively. Furthermore, it is shown that even building from mixed models trained on standard data unrelated to the newly added training and test data can lead to significantly improved recognition results.

  @article{reul2017transfer, abstract = {A method is presented that significantly reduces the character error rates for OCR text obtained from OCRopus models trained on early printed books when only small amounts of diplomatic transcriptions are available. This is achieved by building from already existing models during training instead of starting from scratch. To overcome the discrepancies between the set of characters of the pretrained model and the additional ground truth the OCRopus code is adapted to allow for alphabet expansion or reduction. The character set is now capable of flexibly adding and deleting characters from the pretrained alphabet when an existing model is loaded. For our experiments we use a self-trained mixed model on early Latin prints and the two standard OCRopus models on modern English and German Fraktur texts. The evaluation on seven early printed books showed that training from the Latin mixed model reduces the average amount of errors by 43% and 26%, compared to training from scratch with 60 and 150 lines of ground truth, respectively. Furthermore, it is shown that even building from mixed models trained on standard data unrelated to the newly added training and test data can lead to significantly improved recognition results.}, author = {Reul, Christian and Wick, Christoph and Springmann, Uwe and Puppe, Frank}, journal = {027.7 Journal for Library Culture}, keywords = {sys:relevantfor:uniwue_info6}, title = {Transfer Learning for OCRopus Model Training on Early Printed Books}, year = 2017 }
• LAREX – A semi-automatic open-source Tool for Layout Analysis and Region Extraction on Early Printed Books. Reul, Christian; Springmann, Uwe; Puppe, Frank. In 2nd International Conference on Digital Access to Textual Cultural Heritage (DATeCH). 2017.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  A semi-automatic open-source tool for layout analysis on early printed books is presented. LAREX uses a rule based connected components approach which is very fast, easily comprehensible for the user and allows an intuitive manual correction if necessary. The PageXML format is used to support integration into existing OCR workflows. Evaluations showed that LAREX provides an efficient and flexible way to segment pages of early printed books.

  @article{reul2017larex, abstract = {A semi-automatic open-source tool for layout analysis on early printed books is presented. LAREX uses a rule based connected components approach which is very fast, easily comprehensible for the user and allows an intuitive manual correction if necessary. The PageXML format is used to support integration into existing OCR workflows. Evaluations showed that LAREX provides an efficient and flexible way to segment pages of early printed books.}, author = {Reul, Christian and Springmann, Uwe and Puppe, Frank}, journal = {2nd International Conference on Digital Access to Textual Cultural Heritage (DATeCH)}, keywords = {incunabula}, title = {LAREX – A semi-automatic open-source Tool for Layout Analysis and Region Extraction on Early Printed Books}, year = 2017 }
• Case Study of a highly automated Layout Analysis and OCR of an incunabulum: ‘Der Heiligen Leben’ (1488). Reul, Christian; Dittrich, Marco; Gruner, Martin. In 2nd International Conference on Digital Access to Textual Cultural Heritage (DATeCH). 2017.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  This paper provides the first thorough documentation of a high quality digitization process applied to an early printed book from the incunabulum period (1450-1500). The entire OCR related workflow including preprocessing, layout analysis and text recognition is illustrated in detail using the example of ‘Der Heiligen Leben’, printed in Nuremberg in 1488. For each step the required time expenditure was recorded. The character recognition yielded excellent results both on character (97.57%) and word (92.19%) level. Furthermore, a comparison of a highly automated (LAREX) and a manual (Aletheia) method for layout analysis was performed. By considerably automating the segmentation the required human effort was reduced significantly from over 100 hours to less than six hours, resulting in only a slight drop in OCR accuracy. Realistic estimates for the human effort necessary for full text extraction from incunabula can be derived from this study. The printed pages of the complete work together with the OCR result is available online 1 ready to be inspected and downloaded.

  @article{reul2017study, abstract = {This paper provides the first thorough documentation of a high quality digitization process applied to an early printed book from the incunabulum period (1450-1500). The entire OCR related workflow including preprocessing, layout analysis and text recognition is illustrated in detail using the example of ‘Der Heiligen Leben’, printed in Nuremberg in 1488. For each step the required time expenditure was recorded. The character recognition yielded excellent results both on character (97.57%) and word (92.19%) level. Furthermore, a comparison of a highly automated (LAREX) and a manual (Aletheia) method for layout analysis was performed. By considerably automating the segmentation the required human effort was reduced significantly from over 100 hours to less than six hours, resulting in only a slight drop in OCR accuracy. Realistic estimates for the human effort necessary for full text extraction from incunabula can be derived from this study. The printed pages of the complete work together with the OCR result is available online 1 ready to be inspected and downloaded.}, author = {Reul, Christian and Dittrich, Marco and Gruner, Martin}, journal = {2nd International Conference on Digital Access to Textual Cultural Heritage (DATeCH)}, keywords = {incunabula}, title = {Case Study of a highly automated Layout Analysis and OCR of an incunabulum: ‘Der Heiligen Leben’ (1488)}, year = 2017 }

• Cross Dataset Evaluation of Feature Extraction Techniques for Leaf Classification. Reul, Christian; Toepfer, Martin; Puppe, Frank. In International Journal of Artificial Intelligence & Applications. 2016.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  In this work feature extraction techniques for leaf classification are evaluated in a cross dataset scenario. First, a leaf identification system consisting of six feature classes is described and tested on five established publicly available datasets by using standard evaluation procedures within the datasets. Afterwards, the performance of the developed system is evaluated in the much more challenging scenario of cross dataset evaluation. Finally, a new dataset is introduced as well as a web service, which allows to identify leaves both photographed on paper and when still attached to the tree. While the results obtained during classification within a dataset come close to the state of the art, the classification accuracy in cross dataset evaluation is significantly worse. However, by adjusting the system and taking the top five predictions into consideration very good results of up to 98% are achieved. It is shown that this difference is down to the ineffectiveness of certain feature classes as well as the increased severity of the task as leaves that grew under different environmental influences can differ significantly not only in colour, but also in shape.

  @article{reul2016cross, abstract = {In this work feature extraction techniques for leaf classification are evaluated in a cross dataset scenario. First, a leaf identification system consisting of six feature classes is described and tested on five established publicly available datasets by using standard evaluation procedures within the datasets. Afterwards, the performance of the developed system is evaluated in the much more challenging scenario of cross dataset evaluation. Finally, a new dataset is introduced as well as a web service, which allows to identify leaves both photographed on paper and when still attached to the tree. While the results obtained during classification within a dataset come close to the state of the art, the classification accuracy in cross dataset evaluation is significantly worse. However, by adjusting the system and taking the top five predictions into consideration very good results of up to 98% are achieved. It is shown that this difference is down to the ineffectiveness of certain feature classes as well as the increased severity of the task as leaves that grew under different environmental influences can differ significantly not only in colour, but also in shape.}, author = {Reul, Christian and Toepfer, Martin and Puppe, Frank}, journal = {International Journal of Artificial Intelligence & Applications}, keywords = {HOCS_features}, title = {Cross Dataset Evaluation of Feature Extraction Techniques for Leaf Classification}, year = 2016 }
• Expectation-driven Text Extraction from Medical Ultrasound Images. Reul, Christian; Köberle, Philipp; Üçeyler, Nurcan; Puppe, Frank. In Studies in Health Technology and Informatics. 2016.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  In this study an expectation-driven approach is proposed to extract data stored as pixel structures in medical ultrasound images. Prior knowledge about certain properties like the position of the text and its background and foreground grayscale values is utilized. Several open source Java libraries are used to pre-process the image and extract the textual information. The results are presented in an Excel table together with the outcome of several consistency checks. After manually correcting potential errors, the outcome is automatically stored in the main database. The proposed system yielded excellent results, reaching an accuracy of 99.94% and reducing the necessary human effort to a minimum.

  @article{reul2016expectationdriven, abstract = {In this study an expectation-driven approach is proposed to extract data stored as pixel structures in medical ultrasound images. Prior knowledge about certain properties like the position of the text and its background and foreground grayscale values is utilized. Several open source Java libraries are used to pre-process the image and extract the textual information. The results are presented in an Excel table together with the outcome of several consistency checks. After manually correcting potential errors, the outcome is automatically stored in the main database. The proposed system yielded excellent results, reaching an accuracy of 99.94% and reducing the necessary human effort to a minimum.}, author = {Reul, Christian and Köberle, Philipp and Üçeyler, Nurcan and Puppe, Frank}, journal = {Studies in Health Technology and Informatics}, keywords = {Text_Extraction}, title = {Expectation-driven Text Extraction from Medical Ultrasound Images}, year = 2016 }
• Autonomous Quadrocopter for Search, Count and Localization of Objects. Gageik, Nils; Reul, Christian; Montenegro, Sergio. In Recent Advances in Robotic Systems. 2016.
  • [ Abstract ]
  • [ BibTeX ]
  • [ URL ]
  This chapter describes and evaluates the design and implementation of a new fully autonomous quadrocopter, which is capable of self-reliant search, count and localization of a predefined object on the ground inside a room. In a preliminary calibration scan the parameters of the object are defined: As an example object a red ball is used. The scan determines the colour and radius of the ball. The implementation and principles of the object recognition and search will be described in detail. After determining the scanning parameters, the autonomous search can be executed. This is done autonomously by the quadrocopter, which uses inertial, infrared, ultrasonic, pressure sensors and an optical flow sensor to determine and control its orientation and position in 6 DOF (degree of freedom). Furthermore the quadrocopter can be equipped with sensors for obstacle detection and collision avoidance such as ultrasonic, infrared, pmd (photo mixing device) and sv (stereo vision) cameras. A camera attached to the quadrocopter and directed at the ground is used to find the searched objects and to determine its positions during the autonomous flight. Hence, objects which fulfil the scanning parameters can be found in different positions. Based on its own known position and the position of the object in the picture of the camera, the position of the detected objects can be determined. Thus repeated detections of objects can be excluded. Consequently, objects can be counted and localized autonomously. The position of the object is transferred to the ground station and compared with the true position to evaluate the system. Two different search situations and two different strategies, breadth first search (BFS) and depth first search (DFS), are investigated and their results are compared. The evaluation shows the potential, constraints and drawbacks of this approach just as the effect of the search strategy, and the most important parameters and indicators such as field of view, masking area and minimal object distance as well as accuracy, performance and completeness of the search. The entire system is composed of low-cost components and constructed from scratch. Its integration in the innovative real-time operating system Rodos developed by the German Aerospace Centre is described in detail. Rodos has been developed for embedded systems such as satellites and comparable aerospace systems.

  @inbook{nils2016autonomous, abstract = {This chapter describes and evaluates the design and implementation of a new fully autonomous quadrocopter, which is capable of self-reliant search, count and localization of a predefined object on the ground inside a room. In a preliminary calibration scan the parameters of the object are defined: As an example object a red ball is used. The scan determines the colour and radius of the ball. The implementation and principles of the object recognition and search will be described in detail. After determining the scanning parameters, the autonomous search can be executed. This is done autonomously by the quadrocopter, which uses inertial, infrared, ultrasonic, pressure sensors and an optical flow sensor to determine and control its orientation and position in 6 DOF (degree of freedom). Furthermore the quadrocopter can be equipped with sensors for obstacle detection and collision avoidance such as ultrasonic, infrared, pmd (photo mixing device) and sv (stereo vision) cameras. A camera attached to the quadrocopter and directed at the ground is used to find the searched objects and to determine its positions during the autonomous flight. Hence, objects which fulfil the scanning parameters can be found in different positions. Based on its own known position and the position of the object in the picture of the camera, the position of the detected objects can be determined. Thus repeated detections of objects can be excluded. Consequently, objects can be counted and localized autonomously. The position of the object is transferred to the ground station and compared with the true position to evaluate the system. Two different search situations and two different strategies, breadth first search (BFS) and depth first search (DFS), are investigated and their results are compared. The evaluation shows the potential, constraints and drawbacks of this approach just as the effect of the search strategy, and the most important parameters and indicators such as field of view, masking area and minimal object distance as well as accuracy, performance and completeness of the search. The entire system is composed of low-cost components and constructed from scratch. Its integration in the innovative real-time operating system Rodos developed by the German Aerospace Centre is described in detail. Rodos has been developed for embedded systems such as satellites and comparable aerospace systems.}, author = {Gageik, Nils and Reul, Christian and Montenegro, Sergio}, booktitle = {Recent Advances in Robotic Systems}, keywords = {Quadrotor}, title = {Autonomous Quadrocopter for Search, Count and Localization of Objects}, year = 2016 }