How to Segment Handwritten Historical Chronicles Using Fully Convolutional Networks?
Agents and Artificial Intelligence, Revised Selected Papers, 13th International Conference, ICAART 2021 (2022)
Abstract
This paper deals with historical document image segmentation with focus on chronicles available in the Porta fontium portal. We build on our previously published database that has precise pixel-level annotations in PAGE format but also utilise other datasets for transfer learning in order to improve the results. We discuss a series of experiments that evaluate possibilities how to train a neural model for image, text and background segmentation. The outcome, in a form of segmentation method with relatively low computational costs and great results, is integrated into the Porta fontium portal to improve its possibilities of searching and publication of the documents.
Authors
BibTex
@InProceedings{10.1007/978-3-031-10161-8_9, author="Baloun, Josef and Kr{\'a}l, Pavel and Lenc, Ladislav", editor="Rocha, Ana Paula and Steels, Luc and van den Herik, Jaap", title="How to Segment Handwritten Historical Chronicles Using Fully Convolutional Networks?", booktitle="Agents and Artificial Intelligence", year="2022", publisher="Springer International Publishing", address="Cham", pages="181--196", abstract="This paper deals with historical document image segmentation with focus on chronicles available in the Porta fontium portal. We build on our previously published database that has precise pixel-level annotations in PAGE format but also utilise other datasets for transfer learning in order to improve the results. We discuss a series of experiments that evaluate possibilities how to train a neural model for image, text and background segmentation. The outcome, in a form of segmentation method with relatively low computational costs and great results, is integrated into the Porta fontium portal to improve its possibilities of searching and publication of the documents.", isbn="978-3-031-10161-8" }
