Unlocking history through automated virtual unfolding of sealed documents imaged by X-ray microtomography

Letterlocking, the centuries-old practice of folding and securing letters to create self-contained envelopes, played a crucial role in global communication security before modern envelopes. While the written content is of primary interest, the folding techniques themselves—revealed through crease marks and seals on opened letters—represent a rich dataset for understanding historical communication security. This study examines 250,000 historical letters to systematize letterlocking techniques, revealing built-in tamper-evident mechanisms that deter interception. However, existing analysis is limited by the destructive practice of opening sealed letters. This research proposes a conservation-focused computational approach using high-resolution X-ray microtomography (XMT) scans to virtually unfold unopened letters, preserving their physical integrity while accessing their contents and letterlocking mechanisms. This non-destructive method allows for the study of authentic, sealed letterpackets, offering a deeper understanding of historical cryptography and communication practices.

Previous research on computational flattening algorithms has successfully addressed the virtual unfolding of scrolls, books, and documents with simple folds. However, these methods have not been applied to the complex folds, tucks, and slits found in letterlocking. Existing studies on letterlocking often rely on the analysis of opened letters, which compromises the original artifact and the subtle material evidence of the locking mechanisms. This work builds upon previous advancements in XMT scanning and digital restoration techniques to create a robust and automated virtual unfolding system that can handle the complex three-dimensional structures found in letterlocked documents.

The researchers developed a fully automated computational pipeline for virtual unfolding of letterpackets. This pipeline consists of five main stages: 1. **XMT Scanning:** High-contrast time delay integration X-ray microtomography (XMT) generates a volumetric dataset representing material density. Inks containing higher Z elements appear as bright regions. 2. **Segmentation:** This step identifies layers of writing substrate, separating them from the surrounding environment using feature point detection based on ridge detection algorithms in 2D and 3D image data. The algorithm estimates substrate thickness and removes points exceeding the maximum expected thickness, which could indicate artifacts. The remaining points are linked to form a 3D mesh. 3. **Flattening:** This stage calculates a distortion-minimizing 2D embedding of the 3D mesh, representing the unfolded state of the letter. The algorithm maps mesh vertices from 3D space to 2D space using a seed triangle and incremental vertex addition, minimizing edge length distortion using spring energy minimization. This is performed separately for each connected component. 4. **Hybrid Mesh Propagation:** This step addresses discontinuities in the segmentation by simultaneously repairing the folded and flattened embeddings. It merges flattened connected components into a common reference frame in 2D using a joint 2D/3D optimization that leverages volumetric scan data and the 3D-to-2D mapping to guide the reconstruction. The method exploits the relationship between distances in the 3D and 2D embeddings to guide mesh repair and merging. 5. **Texturing:** Finally, the pipeline produces a 2D image of the virtually unfolded letter by mapping source voxels from the volumetric scan to their corresponding flattened positions. This creates greyscale images of the letter's contents and crease patterns using mean curvature mapping to visualize fold direction and magnitude. The resolution of these textured images is comparable to a 668 dpi digital image.

The virtual unfolding pipeline was successfully applied to four unopened letterpackets from the Brienne Collection, a rare archive of 17th-century undelivered mail. The researchers were able to reconstruct the contents and crease patterns of each letterpacket, including the successful reading of an unopened letter (DB-1627) for the first time. This letter, dated July 31, 1697, contains a request from Jacques Sennacques to his cousin Pierre Le Pers for a certified copy of a death notice. The analysis of the letterpackets revealed a variety of distinct folding sequences and locking mechanisms. The virtual unfolding results confirmed the researchers' existing letterlocking categorization system, which classifies letterlocking mechanisms based on manipulations (folds/rolls, tucks, slits/holes, adhesive, and locks) and assigns security scores. The study of the Brienne Collection revealed a prevalence of LC5 and LC6 letterlocking categories, indicating a shift towards simpler locking mechanisms towards the end of the 17th century. The developed method demonstrated low geometric distortion when compared to a photograph of an opened letter (DB-2040), validating the accuracy of the virtual unfolding process. The project created a database of the Brienne Collection items, including 1706 opened letters identified as letterlocking examples, categorized by format and security level. This dataset contains 577 unopened letterpackets.

The virtual unfolding pipeline successfully addresses the long-standing challenge of accessing the contents of unopened, historically significant documents without compromising their integrity. This methodology not only allows for the recovery of previously inaccessible textual information but also provides detailed insights into the historical practices of letterlocking and their evolution. The high accuracy and automation of the process suggest broad applicability to various types of historical documents, from scrolls and books to other types of folded or layered artifacts. The findings from the Brienne Collection provide valuable insights into 17th-century communication practices and offer a new lens for understanding historical security systems. The systematization of letterlocking techniques contributes significantly to the field of historical cryptography and material culture studies. The integration of computational methods in cultural heritage research offers a powerful alternative to traditional, destructive analysis methods.

This paper presents a significant advancement in digital humanities research by developing a fully automated virtual unfolding pipeline for unopened letterpackets. The methodology successfully recovers textual information and reveals intricate letterlocking mechanisms without causing damage to the original artifacts. The findings contribute to a better understanding of historical communication security and offer valuable insights into the evolution of letterlocking techniques. Future work could focus on improving the robustness of the pipeline to handle highly warped or damaged documents and explore alternative scanning methods to enhance ink contrast for improved legibility.

The current methodology relies heavily on local interactions between vertices in the mesh, potentially leading to global consistency errors, especially in the presence of scanning artifacts such as those from leaded seals. Inks with densities similar to the writing substrate might show low contrast or be undetectable. The incremental flattening approach might not be suitable for heavily warped or damaged documents. The successful merging of connected components during mesh propagation can be challenging, leading to incomplete or incorrect reconstructions. While addressing these limitations remains a focus for future work, the current implementation demonstrates remarkable potential.