Crema, Stefano; Llena, Manel; Calsamiglia, Aleix; Estrany, Joan; Marchi, Lorenzo; Vericat, Damia; Cavalli, Marco, 2020, Can inpainting improve digital terrain analysis? Comparing techniques for void filling, surface reconstruction and geomorphometric analyses, Earth surface processes and landforms (Print) 45 (2020): 736–755. doi_10.1002/esp.4739,
URL: http://www.cnr.it/prodotto/i/418300

The investigation of form and processes in geomorphology and ecology is highly dependent on topographic data_ a reliable digital terrain representation is in fact a key issue across environmental and earth sciences. In many cases, the processing of high-resolution topographic data (e.g., light detection and ranging (LiDAR), structure from motion) has to face issues such as void filling, vegetation/feature removal and interpolation accuracy that are usually related to (i) intrinsic limitations of the adopted technology, (ii) local conditions affecting the survey or (iii) specific design scenario. In this paper, we develop a methodology to test the accuracy of an image inpainting algorithm to fill data voids in complex mountain areas. The devised experiment exploits the availability of a high-resolution, LiDAR-derived digital terrain model and the inpainting approach accuracy is checked against some widely used interpolation techniques (natural neighbor, spline, inverse distance weighting, kriging). In order to better mimic the actual surface texture, a methodology to introduce local topographic variability to the interpolated surface is also presented. The results show a better performance of the inpainting algorithm especially in the case of complex and rugged topography. Two examples showing an effective usage and accuracy of the proposed technique are reported, highlighting the drawbacks that a poor surface representation can introduce. The whole procedure is made freely available within a Matlab (R) script with the addition of sample files. (c) 2019 John Wiley & Sons, Ltd. (c) 2020 John Wiley & Sons, Ltd.

Data from https://intranet.cnr.it/people/