Border Detection for Seamless Connection of Historical Cadastral Maps

Lenc, Ladislav; Prantl, Martin; Martínek, Jiří; Král, Pavel

doi:10.1007/978-3-030-86198-8_4

Cited by 4 publications

(1 citation statement)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…One important aspect that the methods have in common is the adaptation on small amounts of training data which is common in this domain and hampers larger utilisation of deep learning methods. Lenc et al [5] concentrated on the segmentation of the cadastre border and the detection of important points on it, so-called landmarks. The algorithms mainly relied on FCN networks and the results were post-processed and refined using image processing techniques such as mathematical morphology, skeletonization, etc.…”

Section: Related Workmentioning

confidence: 99%

A System for Automatic Seamless Connection of Historical Cadastral Maps

Lenc,

Baloun,

Martínek

et al. 2024

Preprint

View full text Add to dashboard Cite

This paper presents a way to create a complex system for the automatic connection of individual map sheets into one seamless map. The goal is to provide a seamlessly connected map and allow its online presentation. This way, the maps will be easily available and allow much more efficient work for historians, archivists, and the general public. We also detail a set of methods for processing and analysis of historical cadastral maps and evaluate and compare the methods among them. The processed maps are hand-drawn and bring many challenges for machine processing. We focus on the following tasks: nomenclature detection and recognition; map frame detection; landmark detection, and map area segmentation. The tasks are solved utilising a combination of traditional computer vision techniques and neural networks. The experimental section concentrates on the comparison of different methods and the selection of the best candidates for the final system. Moreover, we have created two annotated datasets that are used for the evaluation of the presented methods. The datasets are publicly available for research purposes.

show abstract

Section: Related Workmentioning

confidence: 99%