SCMAD1.0: An Image Dataset of Shuishu Text Character Library

Drawing upon linguistic research resources such as the Shuishu Character Library, each character entry includes 13 content fields: a unique identifier, Shuishu character image, Chinese meaning, pronunciation (annotated in the International Phonetic Alphabet), annotations, classification (semantic domain), character formation method, etc. This subset is directly applicable to model training and evaluation for supervised learning tasks.Data sources were selected from published works including the Shuishu Character Library, Shuishu (Wedding Volume), Common Character List of Shuishu, and Chinese Shuishu (160-volume facsimile edition).

The dataset was constructed by combining automated pipelines — including PDF scanning, image crawling, OCR-to-Word conversion, and textual information extraction — with manual proofreading, ensuring data accuracy and consistency. Subset 2 was specifically designed with machine-learning-friendly features: (1) a variant-relationship network among characters, linking standard glyphs to their variant forms via the variants field; (2) a hierarchical classification system comprising 11 top-level categories, including celestial bodies and astronomical phenomena, heaven-earth and the Eight Trigrams, birds, beasts, fish, insects, and plants, among others; and (3) a unified image naming convention following the format SC_xxxxx.

The dataset supports a wide range of machine learning tasks, including character recognition, classification, cross-modal retrieval, and variant character analysis. The dataset is released under the CC BY-NC 4.0 license on the Science Data Bank platform, providing a standardized benchmark data resource for the intelligent study of Shuishu characters.