Preserving rare and often damaged Ming Dynasty rank badges, exquisite fabric symbols of hierarchical status, presents significant challenges. To address the critical need for high-fidelity restoration of these intricate textiles, we propose a novel contrastive learning-based few-shot inpainting method leveraging Fourier consistency. This work focuses on overcoming challenges in fabric restoration for cultural heritage digitization and makes three key contributions: the creation of a comprehensive dataset of ancient Chinese fabrics; the incorporation of advanced frequency domain features for enhanced pattern understanding; and the development of an extended perceptual loss function to maintain the semantic integrity of complex designs. Quantitative performance metrics demonstrate the effectiveness of the proposed method, achieving an SSIM of 0.8912 with 95.4M parameters.
Zhang et al. (Fri,) studied this question.