The spatial resolution of a Digital Elevation Model (DEM) plays a crucial role in many practical remote sensing applications. However, it is normally limited by the spatial resolution of the raw input imagery, from which a DEM is derived. One solution to enhance the limited resolution of a DEM during the post-processing, is fusing previously obtained high resolution DEM data. This data-driven approach appears particularly promising, considering the recent success of a deep convolutional network in single image super resolution (SISR). In this paper, we propose a new SISR network that can recover a high resolution DEM. Instead of configuring a single network directly mapping low resolution depth values to high resolution depth values, we propose a new model consisting of 3 subnetworks, i.e. a) extracting feature maps; b) inferring the high frequency details; c) refining the result combining the low resolution input with the details from b). This is similar to LapSRN [1] in that both adopt a Laplacian image pyramid to model the scaling process in SISR. However, the proposed method implements a much deeper subnetworks efficiently with multiple recursive feedback and feedforward connections, and an additional Laplacian of Gaussian (LoG)based loss function help to produce more effective training results. In this research, we also produce a high quality DEM dataset obtained from optical and lidar sensors, from satellites and aircraft respectively, covering different scenes found in remote sensing applications. Our experiments demonstrate that the proposed model performs better than other standard deep SISR models in terms of the training convergence and the Peak Signal to Noise Ratio (PSNR) of a reconstructed DEM.

The marching cubes (MC) is a general method which can construct a surface of an object from its volumetric data generated using a shape from silhouette method. Although MC is efficient and straightforward to implement, a MC surface may have discontinuity even though the volumetric data is continuous. This is because surface construction is more sensitive to image noise than the construction of volumetric data. To address this problem, we propose a surface construction algorithm which aggregates local surfaces constructed by the 3-D convex hull algorithm. Thus, the proposed method initially classifies local convexities from imperfect MC vertices based on sliced volumetric data. Experimental results show that continuous surfaces are obtained from imperfect silhouette images of both convex and nonconvex objects.