We study the problem of aggregating a set of polygons by covering them with disjoint representative regions, thereby inducing a clustering of the polygons. Equivalently, this can be seen as a fence enclosure problem, where the goal is to surround the polygons with a set of closed curves. Our objective is to minimize a weighted sum of the total area and the total perimeter of the regions, which naturally extends other fencing problems and has applications in geographical information systems. Previously, this objective was only studied in a restricted variant, in which the boundary curves of the regions must be selected from a fixed subdivision of the plane. It is natural to ask whether the problem is still tractable if this restriction is removed, allowing output regions to be bounded by arbitrary curves. We provide a positive answer in the form of an algorithm with runtime Õ(n⁴), where n is the number of input vertices. To achieve this, we fully characterize the optimal solutions by showing that their boundaries are composed of input edges and circular arcs of constant radius. Additionally, we consider the parametric problem, where for every weighting factor we seek a solution that is optimal for that trade-off of area and perimeter. We show that 𝒪 combinatorial solutions suffice to describe all optimal solutions across all weighting factors, and provide both an exact algorithm and an approximation scheme. To make the algorithms scalable in practice, we develop engineering techniques that exploit structural properties of the solutions. Our experimental evaluation on real-world data shows linear runtime in practice, even for the parametric variant.

doi:10.4230/LIPIcs.ESA.2026.11

@inproceedings{blank2026arbitraryBicriteriaShapes,
    abstract = {We study the problem of aggregating a set of polygons by covering them with disjoint representative regions, thereby inducing a clustering of the polygons. Equivalently, this can be seen as a fence enclosure problem, where the goal is to surround the polygons with a set of closed curves. Our objective is to minimize a weighted sum of the total area and the total perimeter of the regions, which naturally extends other fencing problems and has applications in geographical information systems. Previously, this objective was only studied in a restricted variant, in which the boundary curves of the regions must be selected from a fixed subdivision of the plane. It is natural to ask whether the problem is still tractable if this restriction is removed, allowing output regions to be bounded by arbitrary curves. We provide a positive answer in the form of an algorithm with runtime Õ(n⁴), where n is the number of input vertices. To achieve this, we fully characterize the optimal solutions by showing that their boundaries are composed of input edges and circular arcs of constant radius. Additionally, we consider the parametric problem, where for every weighting factor we seek a solution that is optimal for that trade-off of area and perimeter. We show that 𝒪 combinatorial solutions suffice to describe all optimal solutions across all weighting factors, and provide both an exact algorithm and an approximation scheme. To make the algorithms scalable in practice, we develop engineering techniques that exploit structural properties of the solutions. Our experimental evaluation on real-world data shows linear runtime in practice, even for the parametric variant.},
    address = {Dagstuhl, Germany},
    author = {Blank, Lotte and Eppstein, David and Haunert, Jan-Henrik and Haverkort, Herman and Kolbe, Benedikt and Mayer, Philip and Mutzel, Petra and Naumann, Alexander and Sauer, Jonas},
    booktitle = {34th Annual European Symposium on Algorithms (ESA 2026)},
    doi = {10.4230/LIPIcs.ESA.2026.11},
    editor = {Bille, Philip and Pettie, Seth and Storandt, Sabine},
    isbn = {978-3-95977-445-1},
    issn = {1868-8969},
    pages = {11:1--11:23},
    publisher = {Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
    series = {Leibniz International Proceedings in Informatics (LIPIcs)},
    title = {Bicriteria Polygon Aggregation with Arbitrary Shapes},
    url = {https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.ESA.2026.11},
    urn = {urn:nbn:de:0030-drops-271472},
    volume = {388},
    year = {2026}
}

Measuring the similarity of polygons is a recurring challenge in numerous disciplines. In this work, we focus on the geoinformation domain, where many methods have been proposed to compare building footprints. However, they share key limitations: they typically only compare polygon boundaries, rely on heuristic alignment procedures, and yield a single global similarity score. As a result, they provide little insight into how local sub-structures correspond between polygons. To address these shortcomings, we propose geomGMD, a geometric graph mapping distance. Applied to polygonal skeletons, such as the medial axis, geomGMD produces both a global similarity value and a mapping of local sub-structures, while remaining invariant under translation and rotation. The method is particularly well-suited for building footprints, which are inherently modular polygons composed of sub-structures such as extensions and courtyards. We prove that geomGMD is a metric and derive a binary linear programming formulation for computing it. To assess its practical applicability, we evaluate geomGMD on real-world building footprint data, benchmarking it against multiple state-of-the art methods, and employing it as a core measure in a clustering task. Our method outperforms existing approaches in capturing local structural correspondences and provides richer, more interpretable similarity information. Despite the theoretical hardness of the underlying optimization problem, our model proves to scale well in practice: of more than 130 million distance computations, only 0.02% exceeded a five-minute time limit.

@article{naumann2026footprintAggEdgeOrientation,
    abstract = {Measuring the similarity of polygons is a recurring challenge in numerous disciplines. In this work, we focus on the geoinformation domain, where many methods have been proposed to compare building footprints. However, they share key limitations: they typically only compare polygon boundaries, rely on heuristic alignment procedures, and yield a single global similarity score. As a result, they provide little insight into how local sub-structures correspond between polygons. To address these shortcomings, we propose geomGMD, a geometric graph mapping distance. Applied to polygonal skeletons, such as the medial axis, geomGMD produces both a global similarity value and a mapping of local sub-structures, while remaining invariant under translation and rotation. The method is particularly well-suited for building footprints, which are inherently modular polygons composed of sub-structures such as extensions and courtyards. We prove that geomGMD is a metric and derive a binary linear programming formulation for computing it. To assess its practical applicability, we evaluate geomGMD on real-world building footprint data, benchmarking it against multiple state-of-the art methods, and employing it as a core measure in a clustering task. Our method outperforms existing approaches in capturing local structural correspondences and provides richer, more interpretable similarity information. Despite the theoretical hardness of the underlying optimization problem, our model proves to scale well in practice: of more than 130 million distance computations, only 0.02% exceeded a five-minute time limit.},
    address = {New York, NY, USA},
    author = {Bülte, Laura and Naumann, Alexander and Schlurmann, Johanna and Haunert, Jan-Henrik and Mutzel, Petra},
    doi = {},
    journal = {SIGSPATIAL '26: Proceedings of the 34th ACM International Conference on Advances in Geographic Information Systems},
    note = {Accepted for Publication.},
    pages = {},
    publisher = {Association for Computing Machinery},
    title = {geomGMD: Translation- and Rotation-Invariant Geometric Graph Mapping Distance for Building Footprint Similarity},
    volume = {},
    year = {2026}
}

The increasing availability of multi-temporal building footprint datasets has created new opportunities for studying urban development over time. A key prerequisite for many tasks in urban analytics and temporal map generalization is the aggregation of building footprints into coherent settlement regions. However, dedicated geometry-based methods for temporal aggregation remain unexplored. In the static, non-temporal setting, the bicriteria shapes framework offers an optimization-based approach: It constructs a subdivision of the plane based on the input building footprints. Then, it adds cells from the subdivision to the input to produce aggregated representative regions that are optimal with respect to a geometric objective function. We extend this framework to the temporal setting. To guide the design and evaluation of temporal methods, we identify four desirable properties: temporal consistency, scale consistency, geometric faithfulness, and efficiency. Based on these criteria, we propose several temporal subdivision schemes, integrate them with bicriteria shapes and analyze their strengths and weaknesses both theoretically and experimentally. Our experimental analysis is performed on real-world datasets from OpenStreetMap and the GHS-OBAT project. In addition, we provide an interactive visualization tool that enables direct exploration and comparison of the resulting temporal aggregations.

doi:10.1145/3841645.3844182

@article{naumann2026footprintAggEdgeOrientation,
    abstract = {The increasing availability of multi-temporal building footprint datasets has created new opportunities for studying urban development over time. A key prerequisite for many tasks in urban analytics and temporal map generalization is the aggregation of building footprints into coherent settlement regions. However, dedicated geometry-based methods for temporal aggregation remain unexplored. In the static, non-temporal setting, the bicriteria shapes framework offers an optimization-based approach: It constructs a subdivision of the plane based on the input building footprints. Then, it adds cells from the subdivision to the input to produce aggregated representative regions that are optimal with respect to a geometric objective function. We extend this framework to the temporal setting. To guide the design and evaluation of temporal methods, we identify four desirable properties: temporal consistency, scale consistency, geometric faithfulness, and efficiency. Based on these criteria, we propose several temporal subdivision schemes, integrate them with bicriteria shapes and analyze their strengths and weaknesses both theoretically and experimentally. Our experimental analysis is performed on real-world datasets from OpenStreetMap and the GHS-OBAT project. In addition, we provide an interactive visualization tool that enables direct exploration and comparison of the resulting temporal aggregations.},
    address = {New York, NY, USA},
    author = {Mayer, Philip and Naumann, Alexander and Roth, Felix and Bonerath, Annika and Sauer, Jonas and Mutzel, Petra and Haunert, Jan-Henrik},
    doi = {10.1145/3841645.3844182},
    journal = {SIGSPATIAL '26: Proceedings of the 34th ACM International Conference on Advances in Geographic Information Systems},
    note = {Accepted for Publication.},
    pages = {},
    publisher = {Association for Computing Machinery},
    title = {Temporally Consistent Aggregation of Building Footprints},
    volume = {},
    year = {2026}
}

Agricultural education is highly experience-based and context-dependent, which limits the effectiveness of conventional teaching methods for conveying practical applications. To support more immersive and applied learning, we introduce two educational escape games on vision-related animal welfare aspects in poultry production. One game is implemented in 2D and uses text, pictures, and simple animations, while the other takes place on a virtual 3D farm that allows free movement and spatial arrangement of riddles and puzzles. Both versions cover the same educational content. A comparative user study with university students showed that both games improved knowledge levels in reference to a conventional lecture baseline. The 2D version achieved slightly higher user acceptance as a tool for university teaching, whereas the 3D version led to better quantitative learning outcomes and superior time efficiency. Both games were perceived as motivating and suitable for agricultural knowledge transfer.

doi:10.1177/23998083261469008

@article{BaltzerGeoGames,
    abstract = {Agricultural education is highly experience-based and context-dependent, which limits the effectiveness of conventional teaching methods for conveying practical applications. To support more immersive and applied learning, we introduce two educational escape games on vision-related animal welfare aspects in poultry production. One game is implemented in 2D and uses text, pictures, and simple animations, while the other takes place on a virtual 3D farm that allows free movement and spatial arrangement of riddles and puzzles. Both versions cover the same educational content. A comparative user study with university students showed that both games improved knowledge levels in reference to a conventional lecture baseline. The 2D version achieved slightly higher user acceptance as a tool for university teaching, whereas the 3D version led to better quantitative learning outcomes and superior time efficiency. Both games were perceived as motivating and suitable for agricultural knowledge transfer.},
    author = {Baltzer, Dorian and Douglas, Shannon and Kaufmann, Falko and Tiemann, Inga and Haunert, Jan-Henrik},
    doi = {10.1177/23998083261469008},
    journal = {Environment and Planning B: Urban Analytics and City Science},
    number = {0},
    pages = {},
    publisher = {Sage Journals},
    title = {Interactive geovisualization in educational escape games: Assessing {2D vs 3D} designs on animal welfare awareness},
    volume = {0},
    year = {2026}
}

The automated analysis of historical documents, particularly maps, has drastically benefited from advances in deep learning and its success across various computer vision applications. However, most deep learning-based methods heavily rely on large amounts of annotated training data, which are typically unavailable for historical maps, especially for those belonging to specific, homogeneous cartographic domains, also known as corpora. Creating high-quality training data suitable for machine learning often takes a significant amount of time and involves extensive manual effort. While synthetic training data can alleviate the scarcity of real-world samples, it often lacks the affinity (realism) and diversity (variation) necessary for effective learning. By transferring the cartographic style of a historical map corpus onto modern vector data, we bootstrap an effectively unlimited number of synthetic historical maps suitable for tasks such as land-cover interpretation of a homogeneous historical map corpus. We propose an automatic deep generative approach and an alternative manual stochastic degradation technique to emulate the visual uncertainty and noise, also known as aleatoric uncertainty, commonly observed in historical map scans. To quantitatively evaluate the effectiveness and applicability of our approach, the bootstrapped training datasets were employed for domain-adaptive semantic segmentation on a homogeneous map corpus using a Self-Constructing Graph Convolutional Network, enabling a comprehensive assessment of the impact of our data bootstrapping methods.

doi:10.1007/s10032-026-00582-w

@article{arzoumanidis2026Bootstrapping,
    abstract = {The automated analysis of historical documents, particularly maps, has drastically benefited from advances in deep learning and its success across various computer vision applications. However, most deep learning-based methods heavily rely on large amounts of annotated training data, which are typically unavailable for historical maps, especially for those belonging to specific, homogeneous cartographic domains, also known as corpora. Creating high-quality training data suitable for machine learning often takes a significant amount of time and involves extensive manual effort. While synthetic training data can alleviate the scarcity of real-world samples, it often lacks the affinity (realism) and diversity (variation) necessary for effective learning. By transferring the cartographic style of a historical map corpus onto modern vector data, we bootstrap an effectively unlimited number of synthetic historical maps suitable for tasks such as land-cover interpretation of a homogeneous historical map corpus. We propose an automatic deep generative approach and an alternative manual stochastic degradation technique to emulate the visual uncertainty and noise, also known as aleatoric uncertainty, commonly observed in historical map scans. To quantitatively evaluate the effectiveness and applicability of our approach, the bootstrapped training datasets were employed for domain-adaptive semantic segmentation on a homogeneous map corpus using a Self-Constructing Graph Convolutional Network, enabling a comprehensive assessment of the impact of our data bootstrapping methods.},
    author = {Arzoumanidis, Lukas and Knechtel, Julius and Haunert, Jan-Henrik and Dehbi, Youness},
    doi = {10.1007/s10032-026-00582-w},
    journal = {International Journal on Document Analysis and Recognition (IJDAR)},
    number = { },
    pages = { },
    title = {Automatic uncertainty-aware synthetic data bootstrapping for historical map segmentation},
    volume = { },
    year = {2026}
}

Soil moisture is a fundamental environmental variable essential for applications in agriculture, drought assessment, hydrological modeling, and climate research. In recent decades, numerous Soil Moisture Content (SMC) datasets have been produced from in-situ measurements, airborne and spaceborne observations, and model simulations. However, information about these datasets is fragmented across disparate repositories, publications, and portals, making data discovery, comparison, and integration a significant challenge for researchers. This study introduces the University of Bonn Soil Moisture Data Catalog (UB-SMDC), a unified, standards-compliant metadata portal designed to overcome this fragmentation. Two primary objectives drove the project: first, centralizing information on publicly available SMC datasets into a single, user-friendly, web-based interface; and second, conducting a comprehensive statistical analysis of SMC datasets to characterize the global SMC data landscape. The UB-SMDC was developed using a structured, three-phase methodology: (1) comprehensive data discovery, (2) catalog design and implementation, and (3) metadata creation and analysis. Built on the open-source GeoNetwork platform, the catalog aligns with the Geographic information, ISO19139 metadata standard, to ensure consistency and interoperability. To harmonize key attributes, a controlled keyword taxonomy was developed to standardize terms for spatial and temporal resolution, measurement depth, data format, processing level, and data version. Each metadata record is enriched with links to initial data sources and access, and includes map-based previews to visualize dataset coverage. By assessing 373 metadata records, the UB-SMDC streamlines the data discovery and selection process, supports transparent and reproducible research, and facilitates cross-dataset comparison and integration.

doi:10.1080/20964471.2026.2652079

@article{Mohseni13042026,
    abstract = {Soil moisture is a fundamental environmental variable essential for applications in agriculture, drought assessment, hydrological modeling, and climate research. In recent decades, numerous Soil Moisture Content (SMC) datasets have been produced from in-situ measurements, airborne and spaceborne observations, and model simulations. However, information about these datasets is fragmented across disparate repositories, publications, and portals, making data discovery, comparison, and integration a significant challenge for researchers. This study introduces the University of Bonn Soil Moisture Data Catalog (UB-SMDC), a unified, standards-compliant metadata portal designed to overcome this fragmentation. Two primary objectives drove the project: first, centralizing information on publicly available SMC datasets into a single, user-friendly, web-based interface; and second, conducting a comprehensive statistical analysis of SMC datasets to characterize the global SMC data landscape. The UB-SMDC was developed using a structured, three-phase methodology: (1) comprehensive data discovery, (2) catalog design and implementation, and (3) metadata creation and analysis. Built on the open-source GeoNetwork platform, the catalog aligns with the Geographic information, ISO19139 metadata standard, to ensure consistency and interoperability. To harmonize key attributes, a controlled keyword taxonomy was developed to standardize terms for spatial and temporal resolution, measurement depth, data format, processing level, and data version. Each metadata record is enriched with links to initial data sources and access, and includes map-based previews to visualize dataset coverage. By assessing 373 metadata records, the UB-SMDC streamlines the data discovery and selection process, supports transparent and reproducible research, and facilitates cross-dataset comparison and integration.},
    author = {Farzane Mohseni and Jan-Henrik Haunert},
    doi = {10.1080/20964471.2026.2652079},
    journal = {Big Earth Data},
    number = {0},
    pages = {1--33},
    publisher = {Taylor \& Francis},
    title = {Publicly available soil moisture datasets: an overview and the {UB-SMDC} portal},
    volume = {0},
    year = {2026}
}

Establishing sustainable agricultural systems while ensuring food security has become a global priority. Meeting this goal requires contributions from different fields of agricultural science, many of which depend on detailed information on crops. Recent advancements in deep learning and the transnational harmonization of administrative data have led to the availability of ever-larger datasets of agricultural field polygons. These datasets, however, vary in quality and level of detail. To achieve synergies between different information sources through data fusion and to evaluate the quality of model outputs, it is essential to efficiently identify correspondences in spatially overlapping datasets. We address this challenge by leveraging a state-of-the-art matching algorithm that we adapt by redesigning its connected-component decomposition to handle large-scale datasets of agricultural field polygons. We demonstrate the algorithm’s suitability through two case studies. First, we show how automatically delineated field polygons can be validated against ground truth in terms of their spatial quality. Second, we explore how two established reference datasets align both thematically and spatially. We discuss the dataset comparisons using different evaluation metrics and provide an interactive map viewer that enables the exploration of spatial patterns of the datasets’ alignment by visualizing matching qualities in the geographic context.

doi:10.1080/17538947.2026.2632420

@article{naumann2026aggMatching,
    abstract = {Establishing sustainable agricultural systems while ensuring food security has become a global priority. Meeting this goal requires contributions from different fields of agricultural science, many of which depend on detailed information on crops. Recent advancements in deep learning and the transnational harmonization of administrative data have led to the availability of ever-larger datasets of agricultural field polygons. These datasets, however, vary in quality and level of detail. To achieve synergies between different information sources through data fusion and to evaluate the quality of model outputs, it is essential to efficiently identify correspondences in spatially overlapping datasets. We address this challenge by leveraging a state-of-the-art matching algorithm that we adapt by redesigning its connected-component decomposition to handle large-scale datasets of agricultural field polygons.
We demonstrate the algorithm’s suitability through two case studies. First, we show how automatically delineated field polygons can be validated against ground truth in terms of their spatial quality. Second, we explore how two established reference datasets align both thematically and spatially.
We discuss the dataset comparisons using different evaluation metrics and provide an interactive map viewer that enables the exploration of spatial patterns of the datasets’ alignment by visualizing matching qualities in the geographic context.},
    author = {Naumann, Alexander and Gedicke, Sven and Haunert, Jan-Henrik},
    doi = {10.1080/17538947.2026.2632420},
    journal = {International Journal of Digital Earth},
    number = {1},
    pages = {2632420},
    title = {A {S}calable {M}atching {A}pproach for the {C}omparison of {A}gricultural {L}and {U}se {M}aps {B}ased on {C}orresponding {F}ield {P}olygons},
    volume = {19},
    year = {2026}
}

Prompted by the rapid advancements in software and hardware, 3D building data for numerous different applications is nowadays often captured via mobile or kinematic laser scanning. However, in contrast to other laser scanning methods, there exist only a few approaches tailored for the planning of a kinematic laser scan survey, and none of them provides an optimality guarantee. Therefore, we propose a novel approach based on Mixed Integer Linear Programming (MILP) to find the optimal trajectory for such a survey. To obtain a high-quality point cloud, we account for scanner-related constraints that influence the quality of the resulting point cloud. Moreover, we enable the introduction of tie points to mitigate the effects of uncertainties in the position estimation that are propagated in the acquired data. In our problem formulation, we aim to find the best tour in a properly weighted graph. For this, we propose two different weight settings to either enable a purely length-based optimization or to increase the redundancy in the measurements by incorporating a Visibility Ratio Factor (VRF) into the objective function. To prove the applicability of our approach for offline panning, we apply our formulation to three different scenarios. In this context, the VRF-based weighting enables a significant speed-up of the solving process while resulting in only slightly prolonged routes. This approach paves the way for applying exact algorithms with an optimality guarantee in the planning process for efficient kinematic laser scanning surveys.

doi:10.5194/isprs-annals-XI-2-2026-109-2026

@article{knechtel2026kinematicScanPlanning,
    abstract = {Prompted by the rapid advancements in software and hardware, 3D building data for numerous different applications is nowadays often captured via mobile or kinematic laser scanning. However, in contrast to other laser scanning methods, there exist only a few approaches tailored for the planning of a kinematic laser scan survey, and none of them provides an optimality guarantee. Therefore, we propose a novel approach based on Mixed Integer Linear Programming (MILP) to find the optimal trajectory for such a survey. To obtain a high-quality point cloud, we account for scanner-related constraints that influence the quality of the resulting point cloud. Moreover, we enable the introduction of tie points to mitigate the effects of uncertainties in the position estimation that are propagated in the acquired data. In our problem formulation, we aim to find the best tour in a properly weighted graph. For this, we propose two different weight settings to either enable a purely length-based optimization or to increase the redundancy in the measurements by incorporating a Visibility Ratio Factor (VRF) into the objective function.
To prove the applicability of our approach for offline panning, we apply our formulation to three different scenarios. In this context, the VRF-based weighting enables a significant speed-up of the solving process while resulting in only slightly prolonged routes. This approach paves the way for applying exact algorithms with an optimality guarantee in the planning process for efficient kinematic laser scanning surveys.},
    author = {Knechtel, Julius and Kordgholiabad, Mohammad and Haunert, Jan-Henrik},
    doi = {10.5194/isprs-annals-XI-2-2026-109-2026},
    journal = {ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences},
    pages = {109--116},
    title = {Optimal Path Planning for Kinematic Laser Scanning},
    volume = {XI-2-2026},
    year = {2026}
}

The aggregation of building footprints is a key task of cartographic generalization, which is an important topic in geoinformation science. It has been approached from various angles, ranging from heuristics and optimization algorithms to machine learning. Given a set of input polygons that represent the building footprints, the task is to generate a set of polygons that provide a coarser representation of the input. The problem has applications in the visualization of settlement areas in small-scale maps, as well as settlement classification and analysis. A popular solution approach is to construct a subdivision of the plane and then build a solution by selecting faces from the subdivision. Often, a triangulation is used for the subdivision. However, this can cause the orientations of the boundary edges in the solution to differ drastically from the input polygons, which leads to a loss of information about the underlying settlement structure. We explore an alternative method that constructs the subdivision by extending the input building edges, thereby automatically preserving their orientations. To make the approach scalable to large instances without substantially decreasing the solution quality, we propose different methods of reducing the complexity of the subdivision. Our experimental evaluation on real-world data shows that our method is able to aggregate towns containing up to ≈ 10 000 building footprints while preserving input edge orientations much better than state-of-the-art methods.

doi:10.5194/isprs-annals-XI-4-2026-153-2026

@article{naumann2026footprintAggEdgeOrientation,
    abstract = {The aggregation of building footprints is a key task of cartographic generalization, which is an important topic in geoinformation science. It has been approached from various angles, ranging from heuristics and optimization algorithms to machine learning. Given a set of input polygons that represent the building footprints, the task is to generate a set of polygons that provide a coarser representation of the input. The problem has applications in the visualization of settlement areas in small-scale maps, as well as settlement classification and analysis. A popular solution approach is to construct a subdivision of the plane and then build a solution by selecting faces from the subdivision. Often, a triangulation is used for the subdivision. However, this can cause the orientations of the boundary edges in the solution to differ drastically from the input polygons, which leads to a loss of information about the underlying settlement structure. We explore an alternative method that constructs the subdivision by extending the input building edges, thereby automatically preserving their orientations. To make the approach scalable to large instances without substantially decreasing the solution quality, we propose different methods of reducing the complexity of the subdivision. Our experimental evaluation on real-world data shows that our method is able to aggregate towns containing up to ≈ 10 000 building footprints while preserving input edge orientations much better than state-of-the-art methods.},
    author = {Naumann, Alexander and Bergé, Samuel and Sauer, Jonas and Haunert, Jan-Henrik},
    doi = {10.5194/isprs-annals-XI-4-2026-153-2026},
    journal = {ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences},
    pages = {153--161},
    title = {Building Footprint Aggregation with Preservation of Edge Orientations},
    volume = {XI-4-2026},
    year = {2026}
}

Historical maps represent an invaluable memory which should be preserved. Such kind of maps are, however, mostly scanned and stored as raster graphics which do not contain semantic information in a machine-readable form. To achieve a machine-readable state, an often expensive human intervention is needed in a fully manual or semi-automatic fashion. An automatic interpretation and a feature extraction is then inevitable for a map digitization and vectorization. Automatic approaches showed more and more convincing and promising results on challenging map corpora avoiding human interaction. This paper deals with the semantic segmentation of historical maps based on Graph Convolutional Networks (GCNs) to capture long-range dependencies between image features. This allows for an extension of the receptive field of Convolutional Neural Networks (CNNs) restricted on local dependencies. A Self-Constructing Graph (SCG) module has been applied to automatically induce the structure of the GCN. We performed experiments revealing promising results where our approach achieved an Mean Intersection over Union (mIoU) of 0.68, outperforming a state-of-the-art CNN dedicated to the semantic segmentation of historical maps.

doi:10.1080/15230406.2025.2468304

@article{arzoumanidis2026semSegHistMaps,
    abstract = {Historical maps represent an invaluable memory which should be preserved. Such kind of maps are, however, mostly scanned and stored as raster graphics which do not contain semantic information in a machine-readable form. To achieve a machine-readable state, an often expensive human intervention is needed in a fully manual or semi-automatic fashion. An automatic interpretation and a feature extraction is then inevitable for a map digitization and vectorization. Automatic approaches showed more and more convincing and promising results on challenging map corpora avoiding human interaction. This paper deals with the semantic segmentation of historical maps based on Graph Convolutional Networks (GCNs) to capture long-range dependencies between image features. This allows for an extension of the receptive field of Convolutional Neural Networks (CNNs) restricted on local dependencies. A Self-Constructing Graph (SCG) module has been applied to automatically induce the structure of the GCN. We performed experiments revealing promising results where our approach achieved an Mean Intersection over Union (mIoU) of 0.68, outperforming a state-of-the-art CNN dedicated to the semantic segmentation of historical maps.},
    author = {Lukas Arzoumanidis and Julius Knechtel and Jan-Henrik Haunert and Youness Dehbi},
    doi = {10.1080/15230406.2025.2468304},
    journal = {Cartography and Geographic Information Science},
    number = {2},
    pages = {177--187},
    title = {Semantic segmentation of historical maps using Self-Constructing Graph Convolutional Networks},
    url = {https://www.tandfonline.com/doi/full/10.1080/15230406.2025.2468304},
    volume = {53},
    year = {2026}
}