| Re: surface fittin [message #42974 is a reply to message #10785] |
Fri, 04 March 2005 15:12  |
mchinand
Messages: 66
Registered: September 1996
|
Member |
|
|
In article <d09orm$u9k$1@news.iucc.ac.il>, Tal <ftal@post.tau.ac.il> wrote:
> hello,
>
> i worked with SFIT for that and used a 6th order polynomial which does a
> good job (when looking at the output image it creates). however, when trying
> to reconstruct that image that i see, by using the required x, y and the
> coefficients, the resulting surface is far from being similar to what the
> surface really is. i managed to fully and exactly reconstruct a 3x3
> Gaussian. so i know that i read the KX coefficients correctly. i also
> defined my x and y variables correctly (100% sure). but is doesn't work for
> my 72x15 pixels surface. KX is a floating point output. it could be that
> when saved as double precision, results will be better. but there is no
> keyword for that.
It seems to work for the example in the online help:
=============================================
X = (FINDGEN(61)/10) # REPLICATE(1,61)
Y = TRANSPOSE(X)
; Evaluate a function at each point:
F = -SIN(2*X) + COS(Y/2)
; Compute a sixth-degree polynomial fit to the function data:
result = SFIT(F, 6, kx=k)
=================================================
Now to get the fit from coefficients, first multiply the x and y arrays by 10 since the fit was
done with array indices and not the x and y values used to create F.
x=x*10.
y=y*10.
Create an array for the fit and then loop through the coefficients
fit=fltarr(61,61)
for i=0,6 do for j=0,6 do fit=fit+k(j,i)*x^i*y^j
Hope this helps,
--Mike
--
Michael Chinander
m-chinander@uchicago.edu
Department of Radiology
University of Chicago
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
