Visual Manuscript Analysis Lab
The Visual Manuscript Analysis Lab (VMA) leverages AI advances in computer vision and pattern recognition for manuscript research, addressing practical questions arising from the exploration of historical documents. It is dedicated to the systematic visual analysis of manuscript images, delving into the multifaceted landscape of historical documents with computational precision. VMA aims to contribute to a nuanced understanding of the past through the application of cutting-edge technology and collaborative interdisciplinary inquiry.
The research in VMA spans various topics within the fields of machine learning and computer vision, primarily within the following research directions:
Handwriting Style Analysis
The style in which marks are written on artefacts by scribes can reveal much about the individuals, the artefacts, and their culture. Leveraging advanced machine learning approaches to analyse these marks automatically can provide valuable support for scholars studying these artefacts. This research direction includes topics such as writer retrieval, script identification, style classification, and palaeographic analysis. By using these advanced techniques, scholars can gain deeper insights into ancient writings and the contexts in which they were created, enhancing our understanding of historical cultures and practices.
Visual Navigation of Digitised Manuscripts
Exploring large collections of digitised manuscripts can be challenging for scholars, hindering their progress in answering research questions. In this research direction, we develop machine learning approaches to enable the visual navigation of digitised manuscripts and the automatic retrieval of relevant samples. This research includes topics such as visual-pattern detection, image clustering, and vision-language learning. By employing these advanced techniques, scholars can more effectively navigate and analyse vast digital archives, facilitating a deeper understanding and more efficient examination of historical texts.
Computational Restoration of Digitised Manuscripts
Written artefacts can suffer various forms of degradation due to preservation conditions or the digitisation process. In some cases, content is intentionally damaged, as seen with palimpsests. This research direction focuses on developing machine learning approaches to restore and reconstruct damaged contents in written artefacts using image information and other imaging sources such as MSI and XRFi. It encompasses topics like generative image inpainting, image registration, and image enhancement techniques. By employing these advanced methods, scholars can recover and study the original content of degraded manuscripts.
Pattern Analysis Software Tools
The study of written artefacts generates an ever-increasing amount of digital data in various forms, from raw images to data produced by advanced acquisition techniques. Manual analysis of this data is time-consuming and prone to human error and bias. Thus, a set of Pattern Analysis Software Tools (PAST) has been developed for the automatic analysis of visual and tabular patterns in this research data. These tools facilitate a more efficient study of written artefacts, enabling scholars to benefit from rapid advancements in computational pattern analysis. Additionally, these tools provide insights derived from the statistical analysis of research data. Each PAST tool is developed and tested in collaboration with experts to ensure its usability and relevance to actual research questions.
Current Members of the Visual Manuscript Analysis (VMA) Lab:
Dr. rer. nat. Hussein Mohammed, head of the lab
Dr. Mahdi Jampour, research associate
Dr. Quang-Vinh Dang, research associate
Hui Xu, student assistant
Former Members:
Eric Werner, MA, research assistant
Current Projects:
Tamil palm-leaf manuscripts:
This project aims to develop a system capable of detecting words written in particular styles and colophon-related text on palm-leaf manuscripts from India. Additionally, the system is expected to identify hand-made binding holes in these manuscripts and estimate the distances between them. The images and annotations are provided by the PLMPI initiative as part of our collaboration.
Analysing Rilke’s notebooks:
The aim of this project is to develop a system capable of automatically analysing the complex layouts found in the notebooks of Rainer Maria Rilke and generating cataloguing information such as text orientations, writing implements, and scripts. All data is provided by the Deutsche Literaturarchiv Marbach (DLA) as part of our collaboration.
Restoring the undertext in Georgian palimpsests:
This project utilises generative image inpainting to remove the overtext and restore the visual features beneath, thereby enhancing the readability of the undertext in Georgian palimpsests. The colour and MSI images are provided by the ERC DeLiCaTe project as part of our collaboration.
Completed Projects:
Homer's Iliad on Papyri:
This project aims to build a system capable of detecting Greek letters on papyri, clustering them hierarchically based on handwriting style, and generating representative models for each cluster. The system's outcome is an automatically generated representations of the various styles in which the letters are handwritten. The images and annotations are provided by the EGRAPSA in Basel University.
Seals on Arabic manuscripts:
The aim of this project is to explore the possibilities and limitations of detecting and recognising different seals in Arabic manuscripts from the Staatsbibliothek zu Berlin (SBB) using very few training examples.
Zhangzhung Nyengyu tsakali:
The aim of this project is to analyse the visual features of the Zhangzhung Nyengyu tsakali collection to detect specific symbols, measure the density of sieve prints from the paper-making process, and examine the handwriting styles of potentially different scribes.