I am not sure I understand what Thin-Plate-Splines are.

I thought it was the name of a regularization technique for b-splines surface fitting (i.e. approximation smoothing, not exact interpolation). But reading papers, I understand they are a new type of splines (i.e. not b-splines), built out of the linear combination of RBF (Radial Basis Functions) - and the addition of a plane, to offset things.

I am a bit lost. Can anyone shed some light on this?


Splines that are derived from minimizing the integral of squared 2nd derivatives can be called thin-plate splines. Whether the surface is of B-spline basis or composed of radial basis functions does not matter.

However, it does seem conventionally that only surfaces interpolating a given set of points (while minimizing the integral of squared 2nd derivatives) is referred as thin-plate splines. For surface fitting from minimizing a weighted sum of least squared error and the integral of squared 2nd derivatives, they are typically not referred as thin-plate splines probably because the least squared error term is typically the dominant error to be minimized.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.