Least squares scattered data fitting by truncated SVDs Original Research Article

The l2 fitting of scattered data by tensor products of B-splines in two and three dimensions is considered. A truncated Singular Value decomposition approach is used to automatically account for singularity and ill-conditioning of the Grammian matrix. The method does not require re-gridding and can be applied to data with large holes and irregular footprint, provided that the domain of evaluation is limited to the region where there is enough data. A direct and two iterative methods are discussed and numerical results are offered for several test data sets.