[Insight-users] Phase correlation registration

[Jakub Bican]

Yes, i was presenting 3D PCM at VISAPP09 :))  (and early version at CAIP07)

Back to your issue: there is a problem with any approach that is
affected by the size of overlaps ... this may be even the peak height
as there are more disturbances around the "matching image features" in
the larger overlaps - you'll see.

Just a note to the zero-padding approach: it is only neccessary to
increase the size of the image and fill the new area with zeros - it
is not neccessary to shift the original image to the center of the
resized area. (due to the periodicity of PCM)

Please, let me know about your progress ... or let me fix any problems
with my ITK PCM implementation.

Thanks. Regards, Jakub


2009/2/25 Felix Bollenbeck <Felix.Bollenbeck at iff.fraunhofer.de>:
> Dear Jakub,
>
> thanks for your detailed reply - yes I have tried the same approach using
> the a comparison of all four possible overlaps, with direct intensity
> difference, which didn't work, I will try the using the PC peak height -
> thats a good idea!
>
> I thought also of padding both images to double size, but thats a descent
> computational load.
>
> The wohle thing is a straightforward problem in stitching images for me. And
> direct optimization using the big ITK optimizers is way worse.
>
> I will try both, but I think your suggestion for subimage PC peak height is
> more consistent.
>
>
> Best regards,
>
>    Felix.
>
> P.S.: Yes, I have heard about the 3D PCM, or we actullaly talked about it at
> VISAPP 2009, right?
>
>
>
> -----insight-users-bounces at itk.org wrote: -----
>
> To: Felix.Bollenbeck at iff.fraunhofer.de
> From: Jakub Bican <jakub.bican at matfyz.cz>
> Sent by: insight-users-bounces at itk.org
> Date: 02/24/2009 06:40PM
> cc: insight-users at itk.org
> Subject: Re: [Insight-users] Phase correlation registration
>
> Hi,
>
> the problem is that PCM supposes that the images are periodically
> extended beyond their bounds. This means that shift dX is for PCM the
> same as any shift dX+k*sX where sX is the image size and k is any
> integer.
>
> In the actual code, the shift is computed from the peak using this code:
>
>  for (int i = 0; i < ImageDimension ; i++)
>    {
>    if ( index[i] > vcl_floor( size[i] / 2.0 ) )
>      {
>      offset[i] = -1*(index[i] - size[i]) * spacing[i];
>      }
>    else
>      {
>      offset[i] = -1*index[i] * spacing[i];
>      }
>    }
>
> This approach results in maximum overlap of the two images. If this is
> not enough for you, you have to apply some other metric best suited
> for your images. The simple and probably too expensive aproach would
> be to compute PCM of the overlapped areas for the two cases and
> compare the peak height.
>
> Feel free to ask for any details or materials...
>
> Regards,
>
>    Jakub
>
> PS: PCM submission has never passed IJ compilation and tests execution
> - that is mainly due to problems with fftw and vnl_fft in IJ Testing
> environment. I did several tries to tune the submission before i
> finally gave up. I have to review the latest submission if it
> corresponds to the best available version of ITK PCM code.
>
> PSII: we have also developed a 3D translation and rotation
> registration algorithm based on PCM, but the journal publication of
> the method and IJ publication of ITK code is still under progress :(
> (you can find some conference proceedings by searching for Cylindrical
> Phase Correlation Method)
>
>
> 2009/2/24  <Felix.Bollenbeck at iff.fraunhofer.de>:
>>
>> Hello all,
>>
>> Im currently trying to figure out an elegant heuristic to resolve symmetry
>> (i.e. direction of the shift)
>> of the Phase Correlation for translation registration being implement for
>> ITK.
>>
>> see:
>>
>> http://www.insight-journal.org/browse/publication/138
>>
>>
>> has anyone come accross the symmetry problem, or know an elegant solution?
>>
>> Kind regards,
>>
>>        Felix.
>> _____________________________________
>> Powered by www.kitware.com
>>
>> Visit other Kitware open-source projects at
>> http://www.kitware.com/opensource/opensource.html
>>
>> Please keep messages on-topic and check the ITK FAQ at:
>> http://www.itk.org/Wiki/ITK_FAQ
>>
>> Follow this link to subscribe/unsubscribe:
>> http://www.itk.org/mailman/listinfo/insight-users
>>
>>
> _____________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Please keep messages on-topic and check the ITK FAQ at:
> http://www.itk.org/Wiki/ITK_FAQ
>
> Follow this link to subscribe/unsubscribe:
> http://www.itk.org/mailman/listinfo/insight-users
>
>