Digital elevation models can be considered just like an intensity images. Regardless of working with digital evevation models or digital images, they can be considered as a 2.5 dimensional surface, in which the row, column and grey level values represent the x, y and z respectively, using geometrical representations, such as Delaunay triangulation is possible and even attractive because scaling, translation and rotation operations can be simply implemented by applying affine transformations to the coordinates of the vertices.
The algorithm starts with a few points, a minimum of four points covering the entire area of the image. Then it keeps adding more pixels from the image grid to the triangulation domain to obtain an approximation with smaller error.
Omid B.Sakhi, H. Miar Naeimi, "A Hybrid Refinement/Decimation Triangulation Method for Image Approximation", IET, 2008, 351-356.
S. Soommart and K. Paitoonwattanakij, “Incremental Delaunay Triangulation Algorithm for Digital Terrain Modelling”, Thammasat Int. J. Sc. Tech. , Vol. 4, No.2, July 1999.