ShufaNet: classification method for calligraphers who have reached the professional level

This paper proposes a novel triplet network architecture with a calligraphy-specific attention mechanism for the classification of traditional Chinese character style. The framework addresses the challenge of recognizing highly divergent calligraphic styles through three technical contributions: 1) A hierarchical feature extraction module integrating spatial attention to capture stroke-level characteristics; 2) A style matrix formulation enabling quantitative representation of inter-calligrapher distinctions; 3) A multitask learning strategy that jointly optimizes style classification and similarity metric learning. Experimental results demonstrate that the proposed attention mechanism and style matrix are highly effective for calligraphy style classification, achieving results comparable to those of trained calligraphy experts. Furthermore, the proposed method offers a valuable insight for the study of other calligraphic scripts. The outcomes of this research are expected to advance the development and dissemination of calligraphy and may also serve as a reference for related studies in other fields of traditional culture. Our method achieved a 65% accuracy rate in our data set for few-shot learning, surpassing ResNet and other mainstream CNNs. Meanwhile, we conducted a battle for calligraphy major students, and finally surpassed them. This is the first attempt at deep learning in the field of calligrapher classification, and we hope to provide ideas for subsequent research.