A Format-Character Cooperative Recognition Method for Ancient Chinese Books

This paper proposes a computer-aided collaborative text recognition algorithm for Chinese ancient books, utilizing synergistic techniques in layout analysis and text recognition to effectively address issues of complex formats and image degradation in ancient book texts. First, the DP-LinkNet architecture is used for adaptive binarization, enhancing the clarity of textual content while reducing noise. In the layout analysis, a “character-based column determination“ strategy is employed in conjunction with the DBNet detection algorithm to precisely locate text regions and filter out non-text elements. Next, the text recognition stage applies a “column-based character determination“ strategy using the SVTR LCNet model, which combines a convolutional network with a sparse Transformer mechanism to ensure strong performance in resource-limited environments. Experimental results demonstrate that this method improves character integrity, recognition accuracy, and processing efficiency compared to traditional approaches. This research provides an effective tool for the digital preservation and automated analysis of ancient books, aiding cultural heritage preservation and supporting further academic research.