The generation of 3D models depicting cities in the past holds great potential for documentation and educational purposes. However, it is often hindered by incomplete historical data and the specialized expertise required. To address these challenges, we propose a framework for historical city reconstruction. By integrating procedural modeling techniques and machine learning models within a Geographic Information System (GIS) framework, our pipeline allows for effective management of spatial data and the generation of detailed 3D models. We developed an open-source Python module that fills gaps in 2D GIS datasets and directly generates 3D models up to LOD 2.1 from GIS files. The use of the CityJSON format ensures interoperability and accommodates the specific needs of historical models. A practical case study using footprints of the Old City of Jerusalem between 1840 and 1940 demonstrates the creation, completion, and 3D representation of the dataset, highlighting the versatility and effectiveness of our approach. This research contributes to the accessibility and accuracy of historical city models, providing tools for the generation of informative 3D models. By incorporating machine learning models and maintaining the dynamic nature of the models, we ensure the possibility of supporting ongoing updates and refinement based on newly acquired data. Our procedural modeling methodology offers a streamlined and open-source solution for historical city reconstruction, eliminating the need for additional software and increasing the usability and practicality of the process.

Censuses are structured documents of great value for social and demographic history, which became widespread from the nineteenth century on. However, the plurality of formats and the natural variability of historical data make their extraction arduous and often lead to ungeneric recognition algorithms. We propose an end-to-end processing pipeline, based on optimization, in an attempt to reduce the number of free parameters. The layout analysis is based on semantic segmentation using neural networks for a generic recognition of the explicit column structure. The implicit row structure is deduced directly from the position of the text segments. The handwritten text detection is complemented by an intelligent framing method which significantly improves the quality of the HTR. In the end, we propose to combine several post-correction approaches, neural networks, and language models, to further improve the performance. Ultimately, our flexible methods make it possible to accurately detect more than 98% of the columns and 88% of the rows, despite the lack of graphical separator and the diversity of formats. Thanks to various reframing and post-correction strategies, HTR results reach the excellent performance of 3.44% character error rate on these noisy nineteenth century data. In total, more than 18,831 pages were extracted in 72 censuses over a century. This large historical dataset, as well as training data, is made open-access and released along with this article.

The great potential brought by large-scale data in the humanities is still hindered by the time and technicality required for making documents digitally intelligible. Within urban studies, historical cadasters have been hitherto largely under-explored despite their informative value. Powerful and generic technologies, based on neural networks, to automate the vectorization of historical maps have recently become available. However, the transfer of these technologies is hampered by the scarcity of interdisciplinary exchanges and a lack of practical literature destinated to humanities scholars, especially on the key step of the pipeline: the annotation. In this article, we propose a set of practical recommendations based on empirical findings on document annotation and automatic vectorization, focusing on the example case of historical cadasters. Our recommendations are generic and easily applicable, based on a solid experience on concrete and diverse projects.

Cartography, as a strategic technology, is a historical marker. Maps are tightly connected to the cultural construction of the environment. The increasing availability of digital collections of historical map images provides an unprecedented opportunity to study large map corpora. Corpus linguistics has led to significant advances in understanding how languages change. Research on large map corpora could in turn significantly contribute to understanding cultural and historical changes. We develop a methodology for cartographic stylometry, with an approach inspired by structuralist linguistics, considering maps as visual language systems. As a case study, we focus on a corpus of 10,000 French and Swiss maps, published between 1600 and 1950. Our method is based on the fragmentation of the map image into elementary map units. A fully interpretable feature representation of these units is computed by contrasting maps from different, coherent cartographic series, based on a set of candidate visual features (texture, morphology, graphical load). The resulting representation effectively distinguishes between map series, enabling the elementary units to be grouped into types, whose distribution can be examined over 350 years. The results show that the analyzed maps underwent a steady abstraction process during the 17th and 18th centuries. The 19th century brought a lasting scission between small- and large-scale maps. Macroscopic trends are also highlighted, such as a surge in the production of fine lines, and an increase in map load, that reveal cultural fashion processes and shifts in mapping practices. This initial research demonstrates how cartographic stylometry can be used for exploratory research on visual languages and cultural evolution in large map corpora, opening an effective dialogue with the history of cartography. It also deepens the understanding of cartography by revealing macroscopic phenomena over the long term.

This dataset includes a total of 635 annotated image patches from historical city maps. It is designed for the semantic segmentation of the maps into 3 (map content, road network, background frame) or 5 semantic classes (building blocks, non-built, water, road network, background frame). 330 patches are taken from maps of the city of Paris, while the 305 others are taken from a balanced corpus of city maps from 90 countries all around the world. Please read the detailed informations about data collection methodology, associated metadata and annotation ontology in README.md hereunder :