PRevivor: Reviving Ancient Chinese Paintings using Prior-Guided Color Transformers

TL;DR

This work addresses irreversible color degradation in ancient Chinese paintings by introducing PRevivor, a prior-guided color transformer. The method decouples restoration into luminance enhancement and hue correction, leveraging domain translation to bridge real degraded data with synthetic examples and incorporating localized color priors through a multi-branch hue-correction mechanism. A curated dataset of degraded and well-preserved patches supports rigorous evaluation against six baselines, with results showing superior color fidelity (e.g., improved $ ext{PSNR}$, $ ext{SSIM}$, and reduced $\,\Delta$Colorfulness) and stronger distributional alignment (lower $ ext{FID}$) as well as expert consensus on authenticity. By combining domain-aware luminance restoration and priors-guided hue correction, PRevivor offers a practical pathway toward culturally faithful, scalable digital restoration of heritage paintings, while acknowledging the need for interactive controls and efficiency improvements in future work.

Abstract

Ancient Chinese paintings are a valuable cultural heritage that is damaged by irreversible color degradation. Reviving color-degraded paintings is extraordinarily difficult due to the complex chemistry mechanism. Progress is further slowed by the lack of comprehensive, high-quality datasets, which hampers the creation of end-to-end digital restoration tools. To revive colors, we propose PRevivor, a prior-guided color transformer that learns from recent paintings (e.g., Ming and Qing Dynasty) to restore ancient ones (e.g., Tang and Song Dynasty). To develop PRevivor, we decompose color restoration into two sequential sub-tasks: luminance enhancement and hue correction. For luminance enhancement, we employ two variational U-Nets and a multi-scale mapping module to translate faded luminance into restored counterparts. For hue correction, we design a dual-branch color query module guided by localized hue priors extracted from faded paintings. Specifically, one branch focuses attention on regions guided by masked priors, enforcing localized hue correction, whereas the other branch remains unconstrained to maintain a global reasoning capability. To evaluate PRevivor, we conduct extensive experiments against state-of-the-art colorization methods. The results demonstrate superior performance both quantitatively and qualitatively.

Figures (3)

• Figure 1: Overview of the proposed PRevivor framework. The framework consists of two sequential stages: luminance enhancement and hue correction. The luminance enhancement stage employs domain translation to bridge the gap between real and synthetic degraded data, while the hue correction stage leverages extracted color priors for accurate restoration.
• Figure 2: Qualitative comparison of color restoration results on diverse ancient Chinese paintings. Our method demonstrates superior color fidelity and luminance enhancement while preserving cultural authenticity across various painting themes including flora, portraits, and landscapes. Zoom in for best view.
• Figure 3: Expert preference distribution for top-performing colorization methods. Our method achieves the highest preference scores with strong expert consensus.