On the intelligent construction method of Daqian font library

An intelligent method for constructing a font library is proposed in this article to address the issue of excessive reliance on manual marking and insufficient use of digital technology in the current process of feeding Zhang Daqian’s calligraphy into a font library. With such a method, the deep learning object detection algorithm YOLOv3, known for its ability to extract multi-scale features, is employed for the operation of the automatic detection, positioning, and collection of text images of various sizes from relevant calligraphy works. Furthermore, the deep learning image segmentation algorithm U-Net, featuring a U-shaped structure codec, is utilized to achieve effective separation of foreground details of text strokes from the self-similar background textures in collected images. Subsequently, image vectorization technology is used to achieve the digital conversion of text images. Finally, FontForge, a font editing software, is combined with the aforementioned processes to construct a comprehensive font library comprising 1893 characters. This library provides further digital support for research into Zhang Daqian’s calligraphy.