how does matlab zero pad on fft

hello, 

i am trying to compare my versus matlab's implementaion of a
convolution in the frequency domain.

i have a mask that is not the same size of the image (in fact its much
smaller). but i fft2 the mask with the dimensions of the image and
then convolve with the fft2 of the image.

my question i, what method is used for zero padding the mask to be the
same size of the image?

can someone tell me for example of what the resulting mask would look
like if my original filter was:

-1 0 1
-1 0 1
-1 0 1

and my image size was 128x128? i.e. calling fft2(mask,128,128)

basically i am trying to figure out if the mask stays in the center,
is shoved to the bottom left, or top left etc.

also trying to understand if the mask gets chopped up and centered
around the middle zero and shoved to the bottom left corner.

thank you, happy holidays.

gigabyte@mutimusic.com (CDB) wrote in message news:<be5fe9bf.0311270908.52a447e3@posting.google.com>...
> hello, 
> 
> i am trying to compare my versus matlab's implementaion of a
> convolution in the frequency domain.
> 
> i have a mask that is not the same size of the image (in fact its much
> smaller). but i fft2 the mask with the dimensions of the image and
> then convolve with the fft2 of the image.
> 
> my question i, what method is used for zero padding the mask to be the
> same size of the image?
> 
> can someone tell me for example of what the resulting mask would look
> like if my original filter was:
> 
> -1 0 1
> -1 0 1
> -1 0 1
> 
> and my image size was 128x128? i.e. calling fft2(mask,128,128)
> 
> basically i am trying to figure out if the mask stays in the center,
> is shoved to the bottom left, or top left etc.
> 
> also trying to understand if the mask gets chopped up and centered
> around the middle zero and shoved to the bottom left corner.
> 
> thank you, happy holidays.

If you do like this,

S=real(ifft(fft(ones(3,3),6,6)))  % Note no semicolon!

you will see directly where matlab inserts the zeros.

Rune

"Rune Allnor" <allnor@tele.ntnu.no> wrote in message
news:f56893ae.0311271442.5622995c@posting.google.com...
> gigabyte@mutimusic.com (CDB) wrote in message
news:<be5fe9bf.0311270908.52a447e3@posting.google.com>...
> > hello,
> >
> > i am trying to compare my versus matlab's implementaion of a
> > convolution in the frequency domain.
> >
> > i have a mask that is not the same size of the image (in fact its much
> > smaller). but i fft2 the mask with the dimensions of the image and
> > then convolve with the fft2 of the image.
> >
> > my question i, what method is used for zero padding the mask to be the
> > same size of the image?
> >
> > can someone tell me for example of what the resulting mask would look
> > like if my original filter was:
> >
> > -1 0 1
> > -1 0 1
> > -1 0 1
> >
> > and my image size was 128x128? i.e. calling fft2(mask,128,128)
> >
> > basically i am trying to figure out if the mask stays in the center,
> > is shoved to the bottom left, or top left etc.
> >
> > also trying to understand if the mask gets chopped up and centered
> > around the middle zero and shoved to the bottom left corner.
> >
> > thank you, happy holidays.
>
> If you do like this,
>
> S=real(ifft(fft(ones(3,3),6,6)))  % Note no semicolon!
>
> you will see directly where matlab inserts the zeros.

I didn't think it would insert zeros at all!  I thought the size of the
transform was as specified and that "fft" really means "dft" unless the
dimension is 2^n.

Fred

"Fred Marshall" <fmarshallx@remove_the_x.acm.org> wrote in message news:<1t6dnTHLdJlOGlui4p2dnA@centurytel.net>...
> "Rune Allnor" <allnor@tele.ntnu.no> wrote in message
> news:f56893ae.0311271442.5622995c@posting.google.com...
> > gigabyte@mutimusic.com (CDB) wrote in message
>  news:<be5fe9bf.0311270908.52a447e3@posting.google.com>...
> > > hello,
> > >
> > > i am trying to compare my versus matlab's implementaion of a
> > > convolution in the frequency domain.
> > >
> > > i have a mask that is not the same size of the image (in fact its much
> > > smaller). but i fft2 the mask with the dimensions of the image and
> > > then convolve with the fft2 of the image.
> > >
> > > my question i, what method is used for zero padding the mask to be the
> > > same size of the image?
> > >
> > > can someone tell me for example of what the resulting mask would look
> > > like if my original filter was:
> > >
> > > -1 0 1
> > > -1 0 1
> > > -1 0 1
> > >
> > > and my image size was 128x128? i.e. calling fft2(mask,128,128)
> > >
> > > basically i am trying to figure out if the mask stays in the center,
> > > is shoved to the bottom left, or top left etc.
> > >
> > > also trying to understand if the mask gets chopped up and centered
> > > around the middle zero and shoved to the bottom left corner.
> > >
> > > thank you, happy holidays.
> >
> > If you do like this,
> >
> > S=real(ifft(fft(ones(3,3),6,6)))  % Note no semicolon!
> >
> > you will see directly where matlab inserts the zeros.
> 
> I didn't think it would insert zeros at all!  I thought the size of the
> transform was as specified and that "fft" really means "dft" unless the
> dimension is 2^n.
> 
> Fred

The matlab fft (which, BTW, should be fft2 in the 2D example above) does 
insert zeros if you ask it to. In the example above I made a 3x3 matrix
with unit elements, and ask matlab to zero pad the data to a 6x6 matrix 
before computing the DFT. 

Rune