Automating Iconclass: LLMs and RAG for Large-Scale Classification of Religious Woodcuts

TL;DR

This paper introduces a new method combining Large Language Models and Retrieval-Augmented Generation to classify religious images with high accuracy, significantly improving over traditional search techniques in art history research.

Contribution

It presents a novel hybrid approach using LLMs and vector databases with RAG for large-scale classification of religious woodcuts, leveraging full-page context for improved accuracy.

Findings

Achieved 87% precision at five levels of classification.

Outperformed traditional image and keyword-based search methods.

Demonstrated potential for large-scale analysis in digital humanities.

Abstract

This paper presents a novel methodology for classifying early modern religious images by using Large Language Models (LLMs) and vector databases in combination with Retrieval-Augmented Generation (RAG). The approach leverages the full-page context of book illustrations from the Holy Roman Empire, allowing the LLM to generate detailed descriptions that incorporate both visual and textual elements. These descriptions are then matched to relevant Iconclass codes through a hybrid vector search. This method achieves 87% and 92% precision at five and four levels of classification, significantly outperforming traditional image and keyword-based searches. By employing full-page descriptions and RAG, the system enhances classification accuracy, offering a powerful tool for large-scale analysis of early modern visual archives. This interdisciplinary approach demonstrates the growing potential of LLMs and RAG in advancing research within art history and digital humanities.