Unveiling alternative quantitative narratives of gender and employment in Southern Europe from 19th- and 20th-Century photographs: from pixels to socioeconomic insights using computer vision techniques

Photographs represent a rich yet underutilized source of demographic and economic information, whose integration can move analyses beyond an illustrative role and foster methodological innovation in historical demography. In this research note, we demonstrate how computer vision techniques can be applied to extract complementary data from non-textual sources, alongside conventional demographic and socioeconomic records such as population registers and censuses, thereby constructing the Photo-BALL database as an extension of the BALL database. For this purpose, we applied image classification, detection, and captioning using deep neural networks. Specifically, from the collection of images preserved in the County Archive of Baix Llobregat covering 1870–2020, we automatically segment regions containing people, which are then classified according to perceived sex and age. Whenever an image depicts a labour activity, it is identified through image captioning, a process that attributes natural language descriptions to photographs. Finally, the year associated with each image is inferred using automated date estimation, which integrates both colorimetric and object recognition features. Combining these three dimensions (sex, age, and activity) with temporal attribution, we constructed a visual census spanning 1870–1970 and compared it with demographic and socioeconomic structures estimated from the BALL database, with particular attention to women’s work. While women’s labour activity is often invisible in official censuses and population registers, which typically recorded only teachers, seamstresses, or domestic workers, in the rare years when female occupation was documented, most women were employed in the textile industry. In contrast, the visual census reveals a diverse presence of women across sectors such as hospitality, healthcare, and sports, highlighting occupational activities that are largely absent from administrative records. In this sense, integrating both types of data provide a more comprehensive understanding of socio-occupational structures, uncovering patterns of labour participation, gendered roles, and household economies that remain invisible in conventional sources.