Thanks Stefan.
2009/5/11 Stéfan van der Walt
> 2009/5/11 Chris Colbert :
> > Does the scipy implementation do this differently? I thought that since
> FFTW
> > support has been dropped, that scipy and numpy use the same routines...
>
> Just to be clear, I was referring to scipy.signal.fftconvolv
2009/5/11 Chris Colbert :
> Does the scipy implementation do this differently? I thought that since FFTW
> support has been dropped, that scipy and numpy use the same routines...
Just to be clear, I was referring to scipy.signal.fftconvolve, not
scipy's FFT (which is the same as NumPy's).
Regards
Hi Chris,
If you have MxN and PxQ signals, you must pad them to shape M+P-1 x
N+Q-1, in order to prevent circular convolution (i.e. values on the
one end sliding back in at the other).
Regards
Stéfan
2009/5/11 Chris Colbert :
> Stefan,
>
> Did I pad my example incorrectly? Both images were upped
Stefan,
Did I pad my example incorrectly? Both images were upped to the larger
nearest power of 2 (256)...
Does the scipy implementation do this differently? I thought that since FFTW
support has been dropped, that scipy and numpy use the same routines...
Thanks!
Chris
2009/5/11 Stéfan van der
Hi Chris
2009/5/11 Chris Colbert :
> When convolving an image with a large kernel, its know that its faster to
> perform the operation as multiplication in the frequency domain. The below
> code example shows that the results of my 2d filtering are shifted from the
> expected value a distance 1/2
Ok, that makes sense.
Thanks Chuck.
On Mon, May 11, 2009 at 2:41 PM, Charles R Harris wrote:
>
>
> On Mon, May 11, 2009 at 9:40 AM, Chris Colbert wrote:
>
>> at least I think this is strange behavior.
>>
>> When convolving an image with a large kernel, its know that its faster to
>> perform t
On Mon, May 11, 2009 at 9:40 AM, Chris Colbert wrote:
> at least I think this is strange behavior.
>
> When convolving an image with a large kernel, its know that its faster to
> perform the operation as multiplication in the frequency domain. The below
> code example shows that the results of my
at least I think this is strange behavior.
When convolving an image with a large kernel, its know that its faster to
perform the operation as multiplication in the frequency domain. The below
code example shows that the results of my 2d filtering are shifted from the
expected value a distance 1/2
