How can we identify the materials used in medieval manuscripts without touching them? In the CHARM project, we are combining traditional codicology with modern computer vision to answer this question.
During the autumn, our intern Li Tian developed a new computational workflow to analyze multispectral images captured with a Dino-Lite microscope. Traditionally, identifying inks (such as distinguishing between iron gall ink and carbon-based ink) relies heavily on the visual experience of the researcher. However, subtle differences in how materials reflect Infrared (IR) or Ultraviolet (UV) light can be hard to quantify with the naked eye. For hundreds of images, analysing each set manually could also be a significant undertaking.
To solve this, we built a custom Machine Learning pipeline using Python. The algorithm automatically isolates the target ink from the parchment background and calculates "spectral scores" based on physics-informed rules. For example, it measures how transparent an ink becomes under IR light, providing an objective metric to classify it as iron gall (transparent) or carbon-based (opaque).
To make this tool accessible to everyone, we have deployed it as an interactive Web Application. Researchers can now simply upload an image and receive an automated material prediction in seconds, without needing to write any code.
Try the tool here:
View the source code:
We hope this open-source tool will help standardise material analysis and support further interdisciplinary discoveries in heritage science.