STAGED REFINED RESTORATION FRAMEWORK OF ARCHITECTURAL HERITAGE DAMAGED EAVES TILE OF REPEATING PATTERNS

Jiale Zhang; Hanbin Luo; Shiping Liu

Abstract:

Eaves tiles, as significant artefacts of Chinese cultural heritage, are valued for their intricate inscriptions, animal motifs, and repeating patterns, which offer critical insights into historic architecture, art, and ideology. However, environmental exposure and human activity degradation often obscure their repetitive and symmetrical structures, complicating direct analysis. The author proposes a staged restoration framework combining geometric feature extraction and image inpainting to address this. Integrating the Hough transform with the Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) algorithm, our method infers missing regions from rubbings without relying on intact sections of the same tile, ensuring global consistency. A geometric property-targeted database enables feature matching and candidate selection for restoration. Our approach outperforms baselines, achieving an average FID of 27.25, DISTS of 0.2135, L0SSIM of 0.8513, and inpainting duration of 5.59e-2 seconds. By integrating rough-fine geometric extraction, pattern line fitting, and structure-guided inpainting, the framework generates diverse restoration candidates with high structural and aesthetic fidelity, serving as a robust tool for cultural heritage research.