Between Art History and Artificial Intelligence: Identity and Anonymity in Portraiture from 1600 to 1800 – IDANOPO

The IDANOPO project aims to use facial recognition, generated by Artificial Intelligence (AI), on art portraits produced in the 17th and 18th century, in order to identify the people portrayed and place them in their historical context. It brings together researchers in art history and digital image analysis to design a new tool for the historical study of portraits and the circulation of models.
It is estimated that around 25% of models in old portraits preserved today are anonymous or misidentified. To overcome this problem, which has until now been insoluble in the human sciences, this project aims to closely combine the historical approach with facial recognition using AI. The use of computer vision will enable the automated analysis of large corpora of images, of which only a small part will be supervised, as expert annotations are particularly time-consuming. The corpora are drawn from the collections of three heritage institutions taking part in the project (Bibliothèque Nationale de France, Musée du Château de Versailles and Musée de la Comédie-Française) and from images in Wikidata.
The aim is to detect similarities on a large scale in order to establish groupings of faces that will provide art historians with new bases for examination and comparison. Our hypothesis is that these groupings will make it possible not only to identify some of the models, but also to gain a better understanding of the artists’ studio practices, where the duplication and copying of successful works was commonplace. This two-pronged examination of the identity of the models and artists’ practices will enable us to revisit the history of early modern portraiture and its socio-economic function in Europe.
From an AI perspective, the objective of this project is to develop advanced graph-based deep learning techniques to identify individuals in historical artistic portraits through facial recognition, despite variations in artistic style, interpretation, and material degradation. By leveraging Graph Neural Networks (GNNs), the project aims to group portraits based on facial similarities, enabling the identification of recurring individuals and revealing patterns of duplication, stylistic influence, and workshop practices in historical portrait production. Additionally, the integration of Spiking Neural Networks (SNNs) within the GNN-based pipeline is expected to enhance computational efficiency and scalability, facilitating the analysis of large-scale image datasets.
To ensure transparency and foster trust in the automated identification process, the project will incorporate Explainable AI (XAI) methods, providing interpretable insights into the decision-making mechanisms of the models. Our hypothesis is that graph-based deep learning models, combined with XAI techniques, can effectively overcome the challenges inherent to historical portrait analysis, offering art historians both accurate results and meaningful explanations to investigate the identity of portrait subjects and the socio-economic functions of portraiture in Europe.
This project, which closely combines the humanities and computer science, also includes a strong digital humanities component, as part of a total open science approach. The results of the project will not only be published in a collective book and articles but will also appear on a website dedicated to the project. Video clips will also be used to address issues currently being debated in society, such as the role of artificial intelligence in humanities research.