| Re: Accurate/fast interpolation [message #55703] |
Tue, 04 September 2007 20:56 |
Craig Markwardt
Messages: 1869
Registered: November 1996
|
Senior Member |
|
|
Steve <f@k.e> writes:
> Does anybody have a suggestion of speedups that might help in the
> following scenario...
>
> In a series of images there is a very small shift between succesive
> frames due to orbital dynamics on a spacecraft.
>
> For each image I can translate all the pixel locations to and from a
> common reference frame. The shift beween adjacent frames is sub pixel
> [typical value about 0.2].
>
> What I am trying at the moment is to set one image as a common
> reference, covert all the others to sub pixel positions on that
> reference frame and then use triangulate and trigrid to interpolate
> image values onto this common reference frame. This seems to work but
> is painfully slow [trigrid is fine triangulate takes many seconds].
>
> I just wondered since my data is nearly on the right grid to start
> with if there were a quicker way to do this?
Why not use INTERPOLATE?
If it's true that the points are very nearly on the right grid
already, then nearest neighbor interpolation can do quite well.
Cubic interpolation with CUBIC= set to some number between -0.5 and
-1.0 should be quite good too. I have a nice paper by Parker Kenyon and
Troxel (1983; IEEE Transactions on Medical Imaging, Vol MI-2, No. 1 p.31)
which compares the various interpolation strategies, and the cubic
methods have better frequency response at various pixel offsets than
bilinear or nearest neighbor (which means less smoothing).
Good luck,
Craig
--
------------------------------------------------------------ --------------
Craig B. Markwardt, Ph.D. EMAIL: craigmnet@REMOVEcow.physics.wisc.edu
Astrophysics, IDL, Finance, Derivatives | Remove "net" for better response
------------------------------------------------------------ --------------
|
|
|
|
| Re: Accurate/fast interpolation [message #55717 is a reply to message #55703] |
Tue, 04 September 2007 07:47  |
mattf
Messages: 13
Registered: January 2007
|
Junior Member |
|
|
On Sep 4, 7:32 am, Steve <f...@k.e> wrote:
> Does anybody have a suggestion of speedups that might help in the
> following scenario...
>
> In a series of images there is a very small shift between succesive
> frames due to orbital dynamics on a spacecraft.
>
> For each image I can translate all the pixel locations to and from a
> common reference frame. The shift beween adjacent frames is sub pixel
> [typical value about 0.2].
>
> What I am trying at the moment is to set one image as a common
> reference, covert all the others to sub pixel positions on that
> reference frame and then use triangulate and trigrid to interpolate
> image values onto this common reference frame. This seems to work but
> is painfully slow [trigrid is fine triangulate takes many seconds].
>
> I just wondered since my data is nearly on the right grid to start with
> if there were a quicker way to do this?
>
> Any help gratefully appreciated
>
> S.R.Crothers [at] rl.ac.uk
Maybe a brute-force approach-- 'up-sample' your data so that the
displacements are full pixels rather than part-pixels. This leaves you
with a quantization error, but you've already got a larger
quantization error in the original image, no?
|
|
|
|
| Re: Accurate/fast interpolation [message #55718 is a reply to message #55717] |
Tue, 04 September 2007 07:47 |
Mike[2]
Messages: 99
Registered: December 2005
|
Member |
|
|
On Sep 4, 7:32 am, Steve <f...@k.e> wrote:
> What I am trying at the moment is to set one image as a common
> reference, covert all the others to sub pixel positions on that
> reference frame and then use triangulate and trigrid to interpolate
> image values onto this common reference frame. This seems to work but
> is painfully slow [trigrid is fine triangulate takes many seconds].
> I just wondered since my data is nearly on the right grid to start with
> if there were a quicker way to do this?
One way is to use interpolate afte transforming the homogeneous
coordinates of the points of your image with the !p.t matrix. Suppose
your image data is in an [Nx,Ny] array. You can set up the !p.t
matrix with something like
saved_pt = !p.t
t3d, /reset
t3d, rotate=[0.0, 0.0, angle], translate=[dx, dy, 0]
matrix = !p.t
!p.t = saved_pt
This sets up !p.t as a transformation matrix for rotation by angle
around the z-axis (or is that -angle?) with translations of dx and dy
along x and y (again, I may be dropping a sign here). Your array of
homogeneous coordinates will be [Nx*Ny,4], call it p0. Then you can
transform those points with the !p.t matrix:
p1 = p0 # matrix
and use the new points (in the rotated and translated coordinate
system) to interpolate the image like this:
new_image = reform(interpolate(image, p1[*, 0], p1[*, 1], p1[*, 2]),
Nx, Ny)
I have never timed this against triangulate and trigrid, but I suspect
it will be faster.
Mike
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
