Re: RBF Interpolation
- From: Bo Schwarzstein <Bo.Schwarzstein@xxxxxxxxx>
- Date: Sun, 12 Apr 2009 19:38:29 -0700 (PDT)
On Apr 10, 4:27 pm, highegg <high...@xxxxxxxxx> wrote:
On Apr 10, 7:10 am, Bo Schwarzstein <Bo.Schwarzst...@xxxxxxxxx> wrote:
Hello everyone,
Is there anybody worked on RBF interpolate the scattering data point
cloud ? Why I could get some negative weights ? I use the gaussian
function \phi = exp(-r*c*c), I found if I modify the "c" from 0.00001
to 20000, sometimes the weights are all positive, that's very strange
to me.
Thansk !
A simple least squares RBF interpolation (which you probably refer to)
does not restruct weights to be positive, they will be negative if
that results in a lower residual. You can restrict the weights to be
positive, but then expect a higher residual (which may still be a
better fit from the practical sense). Using an excessively large c
shrinks the support of the gaussian to essentially a single point,
thus the weights will be almost surely positive if the centers are a
subset of the data, but the interpolant will probably be of no use.
Hello, thanks for your reply. I need to interpolate a 3D particle
system with massive scattered data, so I want to find a fastest method
to interpolate them into a grid. The FastRBF seems could handle this
but it's a commercial software. Could I use the KDTree to found some
neighbors around the position where need to be interpolated, then
apply origin RBF ? Thanks.
.
- References:
- RBF Interpolation
- From: Bo Schwarzstein
- Re: RBF Interpolation
- From: highegg
- RBF Interpolation
- Prev by Date: help....
- Next by Date: Re: help....
- Previous by thread: Re: RBF Interpolation
- Next by thread: Engineering Economic Analysis Solution Manual Newman
- Index(es):
Relevant Pages
|
