Case Study of a highly automated Layout Analysis and OCR of an incunabulum

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.

Indiscapes: Instance Segmentation Networks for Layout Parsing of Historical Indic Manuscripts

“…A number of research groups have invested significant efforts in the creation and maintenance of annotated, publicly available historical manuscript image datasets [1]- [4], [11]- [13]. Other collections contain character-level and wordlevel spatial annotations for South-East Asian palm-leaf manuscripts [5], [10], [14].…”

“…The collection and analysis of historical document images is a key component in the preservation of culture and heritage. Given its importance, a number of active research efforts exist across the world [1]- [6]. In this paper, we focus on palm-leaf and early paper documents from the Indian subcontinent.…”

Improving OCR Accuracy on Early Printed Books by Utilizing Cross Fold Training and Voting

“…The experiment was conducted by following the workflow and example described in section III. Table IV shows the CER achieved on a fixed evaluation set of previously unseen lines from the held-out data of each individual best model (1)(2)(3)(4)(5) and the combined results without (ISRI Voting) and with (Confidence Voting) confidence information. Furthermore, the relative improvement of the combined result with respect to the best/average/worst model is indicated.…”

Transfer Learning for OCRopus Model Training on Early Printed Books

“…32-45. ISSN: 2296-0597 analysis (Reul, Dittrich, and Gruner 2017). 1476 is part of the Early New High German Reference Corpus 11 and 1572 was digitized in order to be added to the AL-Corpus 12 .…”

“…Starting from Breuel et al (2013)'s groundbreaking paper the application of recurrent neural networks with LSTM architecture to the field of OCR of historical printings has made excellent progress (Springmann, Fink, and Schulz 2016;Springmann and Lüdeling 2017;Reul, Dittrich, and Gruner 2017), although it was previously considered nearly impossible for the case of incunabula 1 (Rydberg-Cox 2009). Character accuracy rates (CERs) in the high nineties are now routinely possible for even the earliest printings.…”