Shading-based DEM refinement under a comprehensive imaging model

Abstract

This paper introduces an approach to refine coarse digital elevation models (DEMs) based on the shape-from-shading (SfS) technique using a single image. Different from previous studies, this approach is designed for heterogeneous terrain and derived from a comprehensive (extended) imaging model accounting for the combined effect of atmosphere, reflectance, and shading. To solve this intrinsic ill-posed problem, the least squares method and a subsequent optimization procedure are applied in this approach to estimate the shading component, from which the terrain gradient is recovered with a modified optimization method. Integrating the resultant gradients then yields a refined DEM at the same resolution as the input image. The proposed SfS method is evaluated using 30m Landsat-8 OLI multispectral images and 30m SRTM DEMs. As demonstrated in this paper, the proposed approach is able to reproduce terrain structures with a higher fidelity; and at medium to large up-scale ratios, can achieve elevation accuracy 20–30% better than the conventional interpolation methods. Further, this property is shown to be stable and independent of topographic complexity. With the ever-increasing public availability of satellite images and DEMs, the developed technique is meaningful for global or local DEM product refinement.