A Deep-Learning Model for Edition Identification of Premodern Chinese Rare Books

Abstract

Edition identification is a crucial task in the field of bibliography and history of the book. However, traditional methods need to improve in terms of efficiency and accuracy. This paper uses deep learning technology to construct a more suitable framework for the edition identification of premodern Chinese books. We preprocess the scanned pages of rare books using global threshold binarization and heuristic rule-based methods. Subsequently, we build a deep learning framework based on InceptionResNet-V2 that is better suited for the task of Chinese rare book edition identification. We conducted experiments on eleven categories of datasets from different periods and regions, comparing our method with the state-of-the-art deep network structures. The results show that our constructed framework is effective in the task of Chinese rare book edition identification, with an accuracy rate of 91.3% and a recall rate of 83.4%.