R help - Apr 2010 - Something similar to loess for 2D curves?

Hello,

I have been struggling to find something similar to loess for 2D
curves. I have a number of 2D curves

t -------> (x(t), y(t))

with some noise that I want to filter out. There may also be some
(very few) obvious outliers in the data.

Mind that these are not x -----> y(x) functions, as neither coordinate
is monotone. Because of noise, they might not even be locally
monotone.

I have seen some papers on fitting Bezier curves (possibly several of
them) but I rather use something more loess-like, so that I can:

1) Fit a model to the raw curve.
2) Check the residuals in order to detect possible outliers and
interpolate in those points.

I used two independent loess smoothers on the individual coordinates,
but the results were not satisfactory as the correlation between both
coordinates was ignored.

Any ideas?

Best regards,

Carlos J. Gil Bellosta
http://www.datanalytics.com

Carlos J. Gil Bellosta  <cgb <at> datanalytics.com> writes:
> 
> Hello,
> 
> I have been struggling to find something similar to loess for 2D
> curves. I have a number of 2D curves
> 
> t -------> (x(t), y(t))
> 
> with some noise that I want to filter out. There may also be some
> (very few) obvious outliers in the data.
  thin-plate or tensor product splines in package mgcv?

  Ben Bolker