Daejeon, July 27–31
This project aims to connect digital archives containing extensive collections of photographs taken in Afghanistan during the 1970s, a decade marked by profound socio-political transformation, from a society moving toward modernization to one increasingly destabilized by conflict. The project focuses on two significant collections of historical Afghan imagery.
The first is the
Bibliotheca Afghanica (BA) at Basel University Library, which comprises nearly 150,000 photographs depicting Afghan landscapes, people, cities, architecture, and cultural life.
Bibliotheca Afghanica, see
https://ub.unibas.ch/de/sammlungen/bibliotheca-afghanica/.
The second is the
EWA-76 collection, consisting of roughly 1,500 photographs produced during the Polish ethnological Asian expedition conducted in the summer of 1976 by researchers from Adam Mickiewicz University in Poznań. In the 1980s, 106 black-and-white photographs from this expedition were donated to the BA by expedition member Marek Gawęcki to facilitate broader cultural exchange across the Iron Curtain. The remaining 1,038 black-and-white and 348 color photographs are preserved in the Józef Burszta Digital Archive.
Józef Burszta Digital Archive, see
http://cyfrowearchiwum.amu.edu.pl/archive.
Historical photographs of Afghanistan are dispersed across global institutions, private collections, and locally maintained archives. The BA and EWA-76 holdings represent only two fragments of a broader and fragmented visual record that reflects the region’s complex history. While these images constitute a vital component of Afghanistan’s visual memory, they often remain inaccessible, inconsistently described, and disconnected from their contexts of creation and circulation. This project, therefore, proposes a digital humanities framework for connecting such materials through a knowledge graph (KG) that integrates metadata, image content, and cross-repository relationships. Our approach connects archives without centralizing data, allowing institutions to retain complete control of their collections (Alassi, 2025). Although our initial focus is on the BA and EWA-76 collections, the project’s framework and methodology will be developed in a generalizable manner to support future integration of additional photographic archives.
A dedicated web application will serve as the primary access point for scholars and the public, enabling queries across the distributed network and retrieval of images and associated metadata (see Figure 1). The application will rely on a RESTful API to access the KG stored in a triplestore and retrieve images from participating repositories asynchronously. To construct the RDF-based KG, we will define a provenance-aware ontology that can represent, for each image, its metadata, content, geospatial coordinates, URLs, and provenance (Amman & Alassi, 2024). This open-source ontology will employ the constructs of existing ontologies, such as Dublin Core,
Dublin Core Ontology,
https://www.dublincore.org/resources/glossary/ontology/
. Prov Ontology,
https://www.w3.org/TR/prov-o/. Schema Ontology,
https://schema.org/.
Figure 1: Asynchronous data and image retrieval infrastructure
To enrich the representation of image content, we will develop pattern-recognition tools to classify images into four main groups: landscape, urban areas, people, and women. For this purpose, we will train a three-stage classifier using PyTorch ( Xiaogang, 2016). The first stage will identify the scene type (landscape, urban, mixed/other). The second will detect whether an image contains people. The third, applied only to images containing people, will estimate whether women are visually present. The classification results will be added to each image node in the KG, enabling researchers to identify and compare images belonging to the same categories. This functionality will facilitate case studies on historical change; for example, examining how women’s appearance and presence in public spaces changed over time, or analyzing the transformation of specific urban areas such as Kabul’s bazaar before and after the onset of war. To support these kinds of studies, the platform will include an interactive timeline tool that generates chronologically ordered image sequences based on user-defined search criteria (Alassi, 2024).
Another component of the project focuses on identifying visually similar images, including duplicates, cropped versions, or photographs capturing the same scene with minimal spectral or camera-angle variation (Srivastava et.al, 2021). We will develop a two-step pipeline combining fast candidate detection with precise verification. For robust similarity detection, we will use PyTorch with a pretrained ResNet-50 model
What is ResNet-50?
https://blog.roboflow.com/what-is-resnet-50/
.
For images lacking content descriptions, we will generate short, neutral, multilingual descriptions (French, English, German) by combining existing metadata with visual features extracted by classifiers and processed with large language models such as Llama. To mitigate hallucination risks, we will employ a GraphRAG approach that grounds textual descriptions in the KG (Peng et al., 2024). All automatically generated descriptions will include explicit provenance and confidence scores and will be reviewed and, if necessary, corrected by collaborating experts.
By linking distributed visual archives and augmenting them with transparent, interpretable computational methods, this project seeks to deepen global understanding of Afghanistan’s cultural heritage while supporting local custodianship and scholarly engagement. The resulting multimodal KG will serve not only as a research infrastructure but also as a platform that offers a big-picture view of Afghanistan during its most transformational times. Integrating the hundreds of interviews recorded during EWA-76, along with the expedition’s films and written reports, into this infrastructure will be considered as a subsequent project aimed at linking these rich ethnographic materials with the corresponding photographic evidence. Incorporating these multimodal sources will provide researchers with a more holistic and contextually grounded understanding of the images, enabling deeper analysis of cultural practices, social dynamics, and historical conditions documented during the expedition.