Computer vision for an history of early news photo agencies at scale – HIGH-VISION

High-Vision combines expertise from the human sciences and computer vision in order to address the challenges posed by the mass digitization of heritage photographs. As a result of the historical bias inherent in the training data for computer vision systems used to analyze photographs from the late 19th and early 20th centuries, these systems are not generally suitable for analyzing photographs. This research focuses on the archives of four major press photographic agencies (Forbin, Rol, Bain News and Black Star), developing cutting-edge computer vision techniques and applying multimodal artificial intelligence to historical investigation and computer vision analyses. By combining methodologies, these funds are treated as both quantified digital data and as individual archival and historical documents. The project will thus provide new perspectives on the history of contemporary visual cultures. Early press agencies played a key role in constructing contemporary visualities. It remains difficult, however, to write their history on the industrial scale that characterized the distribution of the images they offered for sale in Europe, the United States, and beyond. As High-Vision aligns methods and tools with the actual amplitude of the observed phenomenon, i.e. the birth of an international visual culture, it helps to understand the historical processes that still shape the perception of the world and events by contemporary societies that are saturated with images. Furthermore, the project will provide institutions with transferable solutions for semi-automated approaches to managing large image collections.