Abstract:Cultural relic digitization is crucial for cultural heritage protection and continuity, with 3D reconstruction and material decomposition serving as key technologies for effectively restoring geometric morphology and diverse material properties. Existing methods often produce blurred geometric reconstruction due to specular reflections and insufficient material decomposition accuracy when handling multi-material relics (e.g., metals, ceramics). To address this issue, this study proposes a 3D reconstruction and material decomposition method for cultural relics based on 2D Gaussian splatting (2DGS). Depth and normal priors from large-scale visual foundation models are fused, and geometric reconstruction is optimized using a multi-constraint loss function to mitigate multi-view consistency failure caused by specular interference. A physically-based rendering (PBR) framework with split-sum approximation is adopted, combined with a two-stage training strategy: Gaussian primitive direct shading initialization and G-buffer deferred shading optimization, which outputs PBR material parameters including albedo, metalness, and roughness, and supports physically consistent high-fidelity relighting applications. Experimental results show that the proposed method outperforms mainstream counterparts in PSNR, SSIM, and LPIPS, while effectively recovering fine geometric details and material characteristics of cultural relics, supporting high-fidelity relighting applications, and providing reliable technical support for cultural relic digitization.