This project develops a computational pipeline for analysing digitised manuscript pages written on stationery with frames and gridlines that were pre-printed from carved woodblocks. The frames printed from different blocks show characteristic features/differences, particularly at their corners, which can serve as a visual fingerprint of the respective layout template. The pipeline combines trained detection models with classical image analysis to locate the frame, detect and recognise these corner features, and assign each page to a template. Because both sides of a sheet were printed from the same block, consecutive scans are analysed in pairs, which resolves cases that a single page leaves ambiguous and makes the identification more robust.

The pipeline then examines what was written on the pre-printed stationery. Text regions are detected, without being transcribed, and each is classified as lying inside the frame, outside it or across its border, since writing that departs from the printed layout template can indicate ‘pages of interest’ that contain, for example, an act associated with the emperor, later content additions. The completeness of each frame corner is also measured, because a corner that is hidden from view may point to a correction made by pasting a slip of paper over the frame. Finally, the detected text regions are compared using simple visual measures, such as stroke width and character size, to indicate whether a page may contain writing by more than one hand. Each step produces tables and annotated images, so that its results can be inspected visually and pages of interest can be selected for closer study.

Hussein Mohammed and Thies Staack

Starting date: April 2026

Relevant project: IRP 18: Compiling and Updating Databases of Judicial Knowledge in Nineteenth-Century China