Off-centers: A new type of Steiner points for computing size-optimal quality-guaranteed Delaunay triangulations

We introduce a new type of Steiner points, called off-centers, as an alternative to circumcenters, to improve the quality of Delaunay triangulations in two dimensions. We propose a new Delaunay refinement algorithm based on iterative insertion of off-centers. We show that this new algorithm has the same quality and size optimality guarantees of the best known refinement algorithms. In practice, however, the new algorithm inserts fewer Steiner points, runs faster, and generates smaller triangulations than the best previous algorithms. Performance improvements are significant especially when user-specified minimum angle is large, e.g., when the smallest angle in the output triangulation is 30^o, the number of Steiner points is reduced by about 40%, while the mesh size is down by about 30%. As a result of its shown benefits, the algorithm described here has already replaced the well-known circumcenter insertion algorithm of Ruppert and has been the default quality triangulation method in the popular meshing software Triangle.