Identification of image circulation by AI in large collections of historical photographs

Project Summary

There is a blind spot for both the humanities and computer science when one considers the mass of heritage photographs now available in online repositories. Historians cannot analyze them on actual scale with their usual tools. Current computer vision and artificial intelligence (CVAI) systems face limitations when dealing with historical images and the visual cultures they reflect: they struggle to “see” the past. As part of a collaboration between LARCA and LIPADE, we propose to combine our expertise to address this gap by pushing the boundaries of historical investigation and research in CVAI for large collections of historical photographs. Early agencies developed in the late 19th century and quickly became major actors of visual culture on a global scale by facilitating the circulation and reproduction of images among illustrated periodicals through standardized practices. Their archives are mostly composed of individual photographic prints and negatives with no adequate written documentation to trace their origins and distribution. The complex nature of these visual and semantic data presents an exciting field of exploration for research in CVAI, with potential for industrial applications. GLAM institutions are actively seeking solutions to enhance their capacity to process the ever-expanding number of visual materials they curate. The project will provide transferable solutions for semi-automated approaches to visual heritage. Advances in image similarity estimation, layout analysis, and automated enrichment of archival metadata will contribute to the growing field of applied CVAI for archives and digital humanities.

Daniel Foliard
daniel.foliard@gmail.com

• Professor at Université Paris Cité

Florence Cloppet (LIPADE)
Camille Kurtz (LIPADE)