Dual-Modal Feature Extraction Method for Bronze Ware Based on Image-Point Cloud Fusion
Zepeng Wang, Fang Yujie, Yang Huijun, Zhiyi Zhang · 2025
Point cloud feature extraction is a pivotal technology in 3D vision, holding great significance in fields such as the digital protection of cultural relics. To address the issue of incomplete feature extraction caused by occlusions and low-texture regions in complex scenes, this paper proposes a bimodal feature collaborative extraction method for bronze cultural relics based on multi-view data. This method integrates 2D image semantics and 3D geometric features to enhance the integrity and accuracy of feature extraction. Firstly, the point cloud is rotated at multiple angles, and spatial constraints are applied to generate multi-view point clouds, which are then projected onto the corresponding two-dimensional images. Secondly, the Canny algorithm is employed to extract the edge features of the two-dimensional projection images. Thirdly, the height difference and principal curvature ratio are utilized to extract the geometric features of the point cloud. Finally, the features of the images and point cloud are fused and mapped, and stable feature points are selected according to a custom strategy. Experimental results demonstrate that the proposed method can stably extract detailed features such as the surface ornaments and inscriptions of bronze artifacts. It is suitable for digital modeling and restoration research of cultural relics, effectively reducing the risk of damage to cultural relics caused by frequent on-site investigations and exhibitions.