imtranslate

[Peter Cloetens]

Søren Hauberg wrote:
> lør, 29 12 2007 kl. 14:06 -0600, skrev Daniel Elliott:
>> Hello,
>>
>> I tried contacting the author, but the email address is no longer
>> functional.  I have not done my homework before asking the question...
>>
>> The imtranslate function takes the absolute value of the image as it
>> is brought out of FFT-space.  Is this because of some property of FFT
>> where it can potentially flip the sign of a pixel?
>>
>> I am using this method to permute images that may have negative
>> values.  Does anyone know why the absolute value is used here?
> I haven't written (or even used) imtranslate so I don't know the
> details. I assume that FFT sometimes returns complex values, so an abs()
> is needed. Depending on your problem, I suggest you either
>   1) Separate the signs from the data and transform each separately.
> Something like:
>     S = sign(im);
>     im = abs(im);
>     St = imtranslate(S, ...)
>     imt = imtranslate(im, ...)
>     T = St.*imt;
>   2) Use the more general 'imperspectivewarp'. You can just apply an
> affine transformation matrix like
>     T = [eye(2), [Tx; Ty]];
> where Tx and Ty are your translations.
> 
> Hope that helps,
> Søren
> 
> _______________________________________________
> Help-octave mailing list
> Help-octave at octave.org
> https://www.cae.wisc.edu/mailman/listinfo/help-octave
> 
Hello,
The succession of fft2/ifft2 may indeed introduce a spurious imaginary 
part one must get rid off. But I think it is much better to use real() 
instead of abs(). In that case it will work for any real image also if 
it contains negative values.
Peter