Wall painting plays an important role in the culture relics of the Mogao Caves in Dunhuang, P.R. China. A novel approach of generating a high-resolution orthoimage of the wall painting is proposed. Since the photographic object is nearly flat and alsothe forward overlap between adjacent images is smaller than 60 percent, the main difficulty to be resolved is the high correlation problem among the unknowns. Improved models of relative orientation and bundle adjustment with virtual constraints have been developed to resolve the high correlation problem. A Voronoi diagram of projective footprints is applied to automatically determine the mosaic lines of ortho-rectified images. The color quality of the generated orthoimage is improved through global minimization of the color differences among overlapped images. The experimental results show that the proposed approach has great potential for conservation of wall paintings with sub-millimeter to millimeter precision.
展开▼
