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

TL;DR

PRevivor is a novel deep learning framework that leverages prior-guided transformers to digitally restore and colorize ancient Chinese paintings, effectively addressing complex degradation issues with high-quality dataset training.

Contribution

The paper introduces PRevivor, a new prior-guided color transformer that decomposes restoration into luminance and hue tasks, utilizing dual-branch modules and localized priors for improved accuracy.

Findings

Outperforms state-of-the-art colorization methods quantitatively.

Achieves superior qualitative restoration results.

Demonstrates effectiveness on paintings from Ming, Qing, Tang, and Song Dynasties.

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.