Abstract:The latest developments in the field of digital humanities have increasingly enabled the construction of large data sets which can be easily accessed and used. These data sets often contain indirect spatial information, such as historical addresses. Historical geocoding is the process of transforming indirect spatial information into direct locations which can be placed on a map, thus allowing for spatial analysis and cross-referencing. There are many geocoders that work efficiently for current addresses. However, these do not tackle temporal information, and usually follow a strict hierarchy (country, city, street, house number, etc.) which is difficult-if not impossible-to use with historical data. Historical data is filled with uncertainty (pertaining to temporal, textual, and positional accuracy, as well as to the reliability of historical sources) which can neither be ignored nor entirely resolved. Our open source, open data, and extensible solution for geocoding is based on extracting a large number of simple gazetteers composed of geohistorical objects, from historical maps. Geocoding a historical address becomes the process of finding one or several geohistorical objects in the gazetteers which best match the historical address searched by the user. The matching criteria are customisable, weighted, and include several dimensions (fuzzy string, fuzzy temporal, level of detail, positional accuracy). Since our goal is to facilitate historical work, we also put forward web-based user interfaces which help geocode (one address or batch mode) and display results over current or historical maps. Geocoded results can then be checked and edited collaboratively (no source is modified). The system was tested on the city of Paris, France, for the 19th and 20th centuries. It showed high response rates and worked quickly enough to be used interactively.

This paper introduces a multi-modal network that learns to retrieve by content vertical aerial images of French urban and rural territories taken about 15 years apart. This means it should be invariant against a big range of changes as the (natural) landscape evolves over time. It leverages the original images and semantically segmented and labeled regions. The core of the method is a Siamese network that learns to extract features from corresponding image pairs across time. These descriptors are discriminative enough, such that a simple kNN classifier on top, suffices as final geo-matching criteria. The method outperformed SOTA "off-the-shelf" image descriptors GEM and ResNet50 on the new aerial images dataset.

No abstract

No abstract

No abstract