The costume colors in Dunhuang murals carry important historical and cultural information, yet long-term weathering and human intervention have caused severe color distortion, making scientific restoration an urgent task. This paper, based on visible spectral analysis, sorts out the visual characteristics of Dunhuang costume images and the commonly used pigments, analyzes their deterioration pathways and corresponding spectral variation mechanisms, and proposes a feasible color restoration framework grounded in precise sampling and data modeling. Taking representative cases from the Mogao Grottoes as examples, the paper clarifies the multiple objectives of color restoration, seeks a balance between “material authenticity” and “visual authenticity”, and establishes a multi-scheme and interpretable mechanism, so as to provide methodological support and practical reference for the digital reproduction and cultural interpretation of Dunhuang costume colors.