Spatial Enhancement of DEM Using Perspective Non-Lambertian SFS from Multi-Spectral Image

Numerous geosciences applications need denser and more accurate Digital Elevation Model data. Photogrammetry, as the main method of producing DEM, sometimes loses its feasibility in the textureless area or in the case where no aerial photography with highly spatial resolution is available. Under this circumstance, interpolation methods are often utilized to generate denser DEM map. However, these methods are mathematical processes which only consider known points and neglect the ground truth of unknown surface. To correctly reconstruct the surface of terrain from sparse ground control points, this paper presents a new DEM densification algorithm, combining perspective non-Lambertian SFS based on spectral information from single satellite image with DEM interpolation. Single multi-spectral image was used to correctly estimate surface reflection properties at first, and surface relative shape was reconstructed based on Oren-Nayar reflection model. Finally the shape from gradient interpolation method was applied to generate a higher-resolution DEM grid. Experiments demonstrate the ability of the proposed algorithm for the enhanced resolution DEM map from the sparse DEM grid and multi-spectral image.