Digital Handling Procedure for Digitalizing and Indexing Ancient Manuscripts

This paper presents a comprehensive digital handling procedure for the digitization and indexing of ancient Sundanese palm-leaf manuscripts. The proposed workflow integrates high-resolution image acquisition, glyph-level segmentation, expert-driven annotation, and TEI-compliant XML indexing to preserve the linguistic and cultural information embedded in historical documents. A hybrid segmentation approach, combining manual and projection-based techniques, was employed, with expert-guided annotation ensuring label accuracy. To evaluate the framework, we constructed a benchmark dataset comprising 3,527 manually annotated glyphs from ten manuscript pages. Experimental results demonstrate strong performance: segmentation accuracy of 90.23% with precision of 93.51% and recall of 96.27%, and inter-annotator agreement of 0.87 (Cohen’s kappa), confirming annotation consistency. Furthermore, a baseline syllable recognition test using MobileNetV2 achieved 88.5% Top-1 accuracy, 94.2% Top-3 accuracy, and a macro F1-score of 0.87. Compared to conventional bounding-box segmentation, our polygon-based approach handles irregular glyph contours better, leading to a more reliable OCR performance. These findings validate the technical feasibility and robustness of the proposed method, particularly for low-resource scripts in complex writing systems. Overall, this study contributes a reproducible workflow that not only supports cultural heritage preservation but also lays the groundwork for large-scale digital archives and machine-readable manuscript corpora.