A Design Framework for 3-D Ceramic Patterns Combining an Optimized 3DGS-PBR Model and Kansei Engineering

Chinese ceramic patterns, as treasured elements of intangible cultural heritage, have long faced three major challenges in the process of mapping onto 3D digital models: geometric misalignment on curved surfaces, visual-perceptual distortion under multi-view observation, and the absence of Kansei feedback. These challenges make it difficult for existing approaches to simultaneously preserve cultural semantic fidelity and meet contemporary aesthetic demands. To address this gap, we propose a Kansei-guided human–AI collaborative design framework integrating generative AI, an optimized 3DGS-PBR model, and Kansei Engineering. First, a generative AI–based multi-view image dataset of ceramic vases featuring traditional Chinese patterns is constructed to support 3D reconstruction. Second, within this model, 3D Gaussian Splatting is optimized through joint geometric-pattern regularization and high-order spherical harmonics appearance refinement, coupled with PBR material baking, directly yielding high-fidelity 3D assets that are geometrically accurate, pattern-faithful, and materially realistic. Additionally, we develop a lightweight Kansei prediction model trained on few-shot user feedback to generate interpretable optimization suggestions for PBR parameters, enabling rapid iterative design refinement. Experiments demonstrate that the proposed method achieves superior co-fidelity in geometry, pattern, and material across diverse ceramic forms, outperforming existing 3D reconstruction models. This work offers a novel theoretical perspective and practical approach for the semantically faithful preservation and Kansei-guided innovative design of traditional Chinese ceramic patterns in the digital era.