HHCS: A Dataset of Cipher Symbol Annotations From Handwritten Historical Encrypted Documents for Machine Learning Tasks

Many historical documents, including military and diplomatic correspondence, were encrypted to protect sensitive information. The study of historical cryptology remains an important research area, as deciphering these sources can reveal new historical insights. Despite progress in automated handwritten document processing, only a few projects have focused on encrypted manuscripts due to their scarcity and the specialized expertise required. This research addresses the automated processing of Central European historical handwritten documents from the early modern period, with a focus on encrypted materials and historical cryptology. In this paper, we introduce a dataset of annotated cipher symbols—the Historical Handwritten Cipher Symbol (HHCS) dataset. The dataset is intended to support machine learning tasks in this domain, particularly the detection and segmentation of objects in the studied documents. The dataset is publicly available to the research community and can be downloaded in several widely used formats. Its suitability was verified through a series of experiments, in which state-of-the-art detection and segmentation models were trained on this dataset and subsequently evaluated in terms of classification and detection performance. Based on this dataset, we developed an automated pipeline for processing historical encrypted documents, covering detection, classification, and transcription of encrypted text from scanned images.