This study proposes an automatic scoring approach for calligraphy images using the deep learning model ResNet. The evaluation criteria are grounded in calligraphic aesthetics, emphasizing stroke techniques and the holistic relationships of character structure. The dataset consists of elementary students’ handwriting samples and corresponding expert ratings. After preprocessing and model training, the system generates predicted scores. Experimental results show that the model achieves a low mean absolute error (MAE = 2.6) and a high quadratic weighted kappa (QWK = 0.898), indicating strong consistency between the automated scoring and expert evaluations. The proposed system provides real-time feedback and reduces teachers’ assessment workload. Future work will expand the dataset and refine the scoring dimensions to enhance the model’s applicability in calligraphy education.

Authors:
Yi-Pei Lin, National Taiwan Normal University, Taiwan

About the Presenter(s)
Yi-Pei Lin is currently an art teacher dedicated to fostering students’ artistic expression through contemporary and technology-enhanced approaches.

See this presentation on the full schedule – Friday Schedule

Bookmark