[MATRIX-SIG] convolution

[MISS. K.L.COLLIER]

I'm trying to calculate the convolution of two arrays (an image and a point
 spread function) using 
Numerical python as follows:

#!/usr/local/bin/python

from Numeric import *
from FFT import *

raw_a = array ( [ 1,2,0,3,4,0,0,0,0 ]  )
raw_a.shape=(3,3)

raw_b = array ( [ -1,1,0,-2,2,0,0,0,0 ]  )
raw_b.shape=(3,3)

freq_2a = fft2d(raw_a)
freq_2b = fft2d(raw_b)

fab=freq_2a*freq_2b

res = inverse_fft2d(fab)
print "res"
print res


but there doesn't seem to be an inverse_fft2d function. Is there one?
Would this work anyway?

Do I have to have the two images the same size, i.e. pad out 
the smallest one with zeros. If so do I do this symmetrically or 
fill the zeros on the rhs and bottom of the matrix?

Does python automatically handle the complex numbers ok in the multiplication 
in the frequency domain?

Other options:
1.	I also notice that there is a convolve function in Numeric but that it 		is only 1D. If my psf is a gaussian i.e. is symmetric, is there a 
	way of 	doing 2 1D convolutions to produce a 2D one?

2. 	I believe it can also be done in matrix form if the image is converted 
	to a vector and the psf converted to a block circulant form. If the psf 	is only 5*5 then the matrix would be 25*25, BUT don't I need to first 
	pad out the psf to the same size as the image (512*512) 
	in which case the matrix would be way too big!

Many questions I know, but does someone know the best way to do it?

Thanks,
Karen Collier c9603085@zeus.ac.hud.uk

_______________
MATRIX-SIG  - SIG on Matrix Math for Python

send messages to: matrix-sig@python.org
administrivia to: matrix-sig-request@python.org
_______________