Holistic Vision Pipeline for Large-Scale Manuscript Analysis and Retrieval

This project aims to enhance digital navigation of multigraphic Written Artefacts (WAs) through three complementary computer vision tasks: layout analysis (automatic detection and classification of page regions such as text blocks, images, and marginalia), visual organisation (clustering and mapping of recurring visual concept regardless of language or script), and text segmentation (precise separation of textual content from decorative backgrounds). All methods leverage advanced machine learning approaches to address the visual complexity of manuscripts. The visual analysis in this proposal is essential for several abstract computer vision tasks that require access to various visual elements in manuscript images. By focusing on these tasks, the project directly supports large-scale clustering and efficient retrieval of complex manuscript images, and fostering collaboration between computer science and manuscript studies experts.