MBA
"MBA" is rather a single algorithm than a library. It is developed for making surfaces from scattered data points. The algorithm works at different levels, and can be used both for small and large point sets. Compared to similar methods is extremely fast. The code is implemented in C++.
MBA  Multilevel BSpline Approximation Library Multiresolution methods based on Bspline refinement for approximating scattered data. The methods are extremely fast compared to many other methods and produce fair results. Given scattered data as input, tensor product Bspline surfaces are produced. The algorithms run in a multiresolutional setting over uniform partitions such that the final surface is composed of a sequence of surfaces at dyadic scales:
f = f_{0} + f_{1} + ... + f_{h,}
where f_{i} is a subset of S_{i} for i = 0, ..., h, and S_{0}, ..., S_{h} is a nested sequence of subspaces of S_{h}. Download Go to the downloads page. 
