A shared degradation-aware and context-guided framework for restoring ancient Chinese inscriptions and murals

Long-term degradation obscures the original appearance of ancient Chinese artworks, impeding visual appreciation and cultural interpretation. We present a shared progressive Degradation-Aware and Context-Guided GAN (DACGG) for restoring Dunhuang murals and ancient Chinese character images. DACGG first produces a coarse completion, and then it performs degradation-aware, context-guided refinement through two dedicated modules, which are a Region-Adaptive Frequency-Guided Dynamic Convolution (RAFG-DC) and a Context-Modulated Affine Coupling (CMAC). RAFG-DC adapts convolutional filtering to frequency-dependent heterogeneous degradation, while CMAC injects neighboring contextual cues to improve structural, textural, and chromatic consistency. The proposed framework is evaluated on CHICC and the public DhMurals1714 benchmark. CHICC is a curated ancient Chinese character dataset using color-domain degradation synthesis to simulate additive contamination and fading degradation. Evaluations on the two datasets demonstrate improved restoration performance over comparative methods, indicating the potential of shared digital restoration strategies to support the conservation of degraded pictorial and inscriptional heritage images.