Nonlocal similarity based DEM super resolution

• Published in: ISPRS journal of photogrammetry and remote sensing Vol. 110; pp. 48 - 54
• Main Authors: Xu, Zekai, Wang, Xuewen, Chen, Zixuan, Xiong, Dongping, Ding, Mingyue, Hou, Wenguang
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.12.2015
• Subjects: Digital elevation model; Nonlocal algorithm; Reconstruction; Super resolution; Reconstruction; Digital elevation model; Nonlocal algorithm; Super resolution
• ISSN: 0924-2716; 1872-8235
• DOI: 10.1016/j.isprsjprs.2015.10.009

This paper discusses a new topic, DEM super resolution, to improve the resolution of an original DEM based on its partial new measurements obtained with high resolution. A nonlocal algorithm is introduced to perform this task. The original DEM was first divided into overlapping patches, which were classified either as “test” or “learning” data depending on whether or not they are related to high resolution measurements. For each test patch, the similar patches in the learning dataset were identified via template matching. Finally, the high resolution DEM of the test patch was restored by the weighted sum of similar patches under the condition that the reconstruction weights were the same in different resolution cases. A key assumption of this strategy is that there are some repeated or similar modes in the original DEM, which is quite common. Experiments were done to demonstrate that we can restore a DEM by preserving the details without introducing artifacts. Statistic analysis was also conducted to show that this method can obtain higher accuracy than traditional interpolation methods.