Combining Multispectral Imaging and Holographic Projection for Non-Contact Color Restoration of Dunhuang Murals: A Review and Conceptual Framework

Abstract

The historically significant murals of the Dunhuang Mogao Grottoes exhibit severe fading and discoloration resulting from over a millennium of physicochemical deterioration. This study proposes a novel conceptual framework for non-contact visual restoration that synergistically integrates multispectral imaging, deep learning, and computer-generated holography (CGH). The proposed workflow begins with multispectral imaging to capture the spectral signatures of residual pigments. Subsequently, deep learning models—specifically, an invertible neural network for color restoration and generative adversarial networks for image inpainting, potentially fine-tuned with domain-specific constraints for cultural heritage—are employed to infer the original colors. In the final stage, CGH algorithms transform the digitally restored image into a phase-only hologram, enabling its projection onto the mural surface. The results demonstrate that multispectral data constitute the essential foundation for color inference, while deep learning enables effective restoration, and optimized CGH facilitates high-fidelity projection. The seamless integration of these technologies thus establishes a feasible perception-computation-presentation pipeline for non-contact visual restoration. Collectively, this work reviews the state of the art and presents an innovative, interdisciplinary paradigm for cultural heritage conservation, offering a safe, controllable alternative with significant potential for both preservation and public engagement.