| Re: dimension problem in matrix multiplication [message #39025] |
Tue, 13 April 2004 14:58 |
Craig Markwardt
Messages: 1869
Registered: November 1996
|
Senior Member |
|
|
Benjamin Hornberger <benjamin.hornberger@stonybrook.edu> writes:
> Hi all,
>
> I have a 3-dimensional data array "image" (n_cols, n_rows, n_data) where
> cols and rows denote pixels in a 2d-image and data are different data
> values for each pixel. Now I would like to perform a matrix
> multiplication so that the data vector (n_data elements) for each pixel
> is multiplied by a 2-dimensional matrix "c".
>
> If I try result = image # c, I get an "incompatible dimensions" error
> since obviously IDL doesn't know that the 3rd dimension in "image" is
> the vector I want to multiply by the matrix, and I want that operation
> done for every pixel. Playing around with * and 0:n_data-1 subscripts
> didn't help so far.
...
Greetings, I think you won't be able to get the '#' or '##' operators
to work very nicely with 3D matrices.
It may be possible to:
1. Do it yourself with FOR loops and TOTAL(), which may not be too
bad if you can slice it right. Example, you can remove your
original inner loop by doing,
cji = rebin(reform(C(*,I), 1, 1, n_data), n_cols, n_rows, n_data)
result(*,*,I) = total(image*cji, 3)
This probably has the fewest total loop iterations. Of course
you need twice the storage of IMAGE.
2. Loop through each pixel, and compute the matrix product explicitly
image_vec = reform(image(I,J,*))
result(I,J,*) = c # image_vec
Where I probably have screwed up the # vs ## thing.
Good luck,
Craig
Craig
--
------------------------------------------------------------ --------------
Craig B. Markwardt, Ph.D. EMAIL: craigmnet@REMOVEcow.physics.wisc.edu
Astrophysics, IDL, Finance, Derivatives | Remove "net" for better response
------------------------------------------------------------ --------------
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
