Automatic Elevation Contour Vectorization: A Case Study in a Deep Learning Approach

Historical maps contain valuable topographic information, including altimetry in the form of annotated elevation contours. These contours are essential for understanding past terrain configurations, particularly in areas affected by human activities such as mining or dam construction. To make this d...

Full description

Saved in:
Bibliographic Details
Main Authors: Jakub Vynikal, Jan Pacina
Format: Article
Language:English
Published: MDPI AG 2025-05-01
Series:ISPRS International Journal of Geo-Information
Subjects:
Online Access:https://www.mdpi.com/2220-9964/14/5/201
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Historical maps contain valuable topographic information, including altimetry in the form of annotated elevation contours. These contours are essential for understanding past terrain configurations, particularly in areas affected by human activities such as mining or dam construction. To make this data usable in modern GIS applications, the contours must be vectorized—a process that often requires extensive manual work due to noise, inconsistent symbology, and topological disruptions like annotations or sheet boundaries. In this study, we apply a convolutional neural network (U-Net) to improve the automation of this vectorization process. Leveraging a recent benchmark for historical map vectorization, our method demonstrates increased robustness to disruptive factors and reduces the need for manual corrections.
ISSN:2220-9964