期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2002
卷号：XXXIV-5/W3
出版社：Copernicus Publications
摘要：Digital Elevation ModelSDEMTinterpolation is one basic functions for spatial description and spatial analysis inGIS and related spatial information fields. DEM interpolation can be viewed as a function for determining theheights of unknown points using a set of proper known data. So, selecting a set of appropriate neighboringreference data points is one of the key steps for DEM interpolation. The selected reference points are used forestimating the value of elevation at any location in the given area. The commonly used search radius and quadrantsearch methods take reference points according to metric distance and are considered as metric methods. Thereference points selected by a metric method might not be well distributed in space. This leads to the discontinuityproblem of interpolated DEM surface and some artifacts might be generated.The use of Voronoi diagrams and 'area-stealing' (or natural neighbor interpolation) has been shown to adapt well topoor data distributions because insertion into the mesh of a sampling point generates a well–defined set ofneighbors. The 'stolen area' of the immediate neighbors (also called first order neighborsTare used fordetermining the weighting function. However, other non-first order neighbors may also have influence orcontribution to the value of estimated elevation because the topographic surface is continuous. In other word, it isrational to take K-order neighbors into account in DEM interpolation.A Voronoi K-order based approach for DEM interpolation is discussed in this paper.The limitations of traditional metric methods and the improvement achieved by Voronoi 'area-stealing' arediscussed in the first part of the paper. The concept and computation of Voronoi k order neighbor are introduced inthe second part. The use of Voronoi k order neighbors for selecting reference points and determining weightingfunctions are discussed in the third part. The k order-based interpolation approach is compared with metric and 1storder approaches in the last section of the paper
