YOLOv8-seg-based instance segmentation for Nanyang Han Dynasty relief stone rubbings and cultural creative design

Abstract

The digitization of Nanyang Han Dynasty relief stone rubbings faces severe obstacles due to inherent features like coarse textures, heavy noise interference, and dense pattern occlusion. To overcome the deficiencies of existing algorithms in handling background complexity and separating adhered patterns, this study proposes an integrated framework for Instance Segmentation and cultural creative design utilizing YOLOv8-Seg. Specifically, a specialized dataset comprising 621 samples was established, incorporating Mosaic data augmentation to enhance model robustness and alleviate issues related to data scarcity. Empirical evaluations reveal that the model achieves a Precision of 98.1% and a mAP50–95 of 79.2% on the test set. These metrics demonstrate superior noise immunity and boundary segmentation accuracy, significantly surpassing the traditional Otsu thresholding method. Furthermore, a comprehensive pipeline linking “intelligent segmentation, vectorization, and cultural creative design” was constructed to facilitate the modern reinterpretation of Han relief patterns. This research contributes a novel technical paradigm for the digital preservation and innovative transformation of cultural heritage.