Revealing latent linear structures in murals via joint analysis of hyperspectral and visible images

As vital components of cultural heritage, murals preserve rich linear structural information that is often obscured by aging and surface contamination. Existing studies have demonstrated the value of hyperspectral imaging for mural analysis, but robust cross-modal registration with visible images and reliable extraction of subtle latent line features remain challenging. This paper proposes a joint hyperspectral–visible analysis method for revealing candidate latent linear structures. The method integrates hyperspectral enhancement, PCA-based representative image generation, edge-domain registration, and edge-corner kernel density differencing. Experiments on the Baisha murals achieved sub-pixel registration accuracy (RMSE < 1 pixel) and revealed latent linear structures that were indistinct in visible images. These results demonstrate the effectiveness of the proposed method for hyperspectral–visible image registration and its potential for assisting mural analysis and conservation.