Yi script character detection in ancient manuscripts is challenging due to limited annotated data and complex text structures, including arbitrary layouts, high character density, and non-standard forms. To address these challenges, we propose CloYiDet, a Yi script character detection method that incorporates a dual-branch transformer (Cloformer) for global-local feature fusion, enhancing feature representation by modeling both global semantic and local details. A text kernel stretching (TKS) is introduced to improve the model’s adaptability to varying aspect ratios and dense layouts. To address the data scarcity, a dedicated handwritten Yi script character dataset was constructed from ancient manuscripts. Experiments show that the proposed method achieves a precision of 99.63% and 98.32% on the Yi Handwritten and the Yi Print Dataset, respectively, demonstrating superior performance in detection tasks.
Ding et al. (Wed,) studied this question.