[Insight-users] Re: SetFixedImageRegion

[Luis Ibanez]

Hi Wei,


Please do the following:


1) Stop writing code.



2) Read the section of resampling from the
    ITK Software Guide.

      http://www.itk.org/ItkSoftwareGuide.pdf

    Section 6.7.1, pdf-pages 199 to 218.



3) Read the Chapter on Image Registration from
    the ITK SoftwareGuide. This is Chapter 8,
    pdf-pages 241 - 328.

    All the Examples available in

        Insight/Examples/Registration

    are explained in this Chapter of the software
    guide. To be more precise, the examples *are*
    the SoftwareGuide.



Please let us know if you have any questions
after reading this material,


    Thanks


      Luis



--------------------
Wei-Lin Bee wrote:

> Hi Luis,
> 
> You're right. My image files come from one modality. Sorry for misleading
> you and thank you for correcting me immediately before I go too far.
> 
> As for this case (roughly 5mm X 20mm overlap between 20mm X 20mm images), 
> Due to lack of generic programming knowledge, although you've already given me
> clear direction and the next step I should follow, however, I don't exactly
> know how to do.
> 
> A.What I learned from API is:
>   In itk::ImageRegistrationMethod, there is SetFixedImageRegion function needed
> to be instantiated?
> 
> B.So I try to get similar code from Insight/Examples/Registration:
> 
> 1.ImageRegistration1~14:
> 
>   registration->SetFixedImageRegion(
> fixedImageReader->GetOutput()->GetBufferedRegion() );
> 
>    or
> 
>   registration->SetFixedImageRegion( 
>        fixedNormalizer->GetOutput()->GetBufferedRegion() );
> 
> 2.MeanSquaresImageMetric1
> 
>   metric->SetFixedImageRegion(fixedImage->GetBufferedRegion());
> 
> 3.MultiResImageRegistration1~2
> 
>   registration->SetFixedImageRegion( 
>        fixedCaster->GetOutput()->GetBufferedRegion() );
> 
> 
> 4.DeformableRegistration6
> 
>   FixedImageType::RegionType fixedRegion = fixedImage->GetBufferedRegion();
>   registration->SetFixedImageRegion( fixedRegion );
> 
> It seems that all of the examples are using full available content of images.
> 
> C. Finally, I seemingly got some hints from [Insight-users] mailing list
> archives:
>  
>    ex.http://public.kitware.com/pipermail/insight-users/2004-March/007391.html
> 
>      metric->SetMovingImageStandardDeviation(1.0);
>      metric->SetFixedImageStandardDeviation(1.0); 
>      metric->SetFixedImageRegion(fixedImg->GetBufferedRegion());
>      metric->SetNumberOfSpatialSamples(1000);
> 
> However, I think asking for a complete and accurate answer should be the best
> way to learn ITK. (much better than aimless copy and paste myself)
> On the other hand, I have wasted too much time(3 weeks)trial error in this test
> sample registration(later, I have to apply ITK in 2D, 3D tissue images from 16
> channel two-photon microscopy). I got go faster.
> 
> 
> 
> Thanks abound !
> 
> 
> Best,
> Tiger
> 
> 
> 
> Quoting Luis Ibanez <luis.ibanez at kitware.com>:
> 
> 
>>Hi Tiger,
>>
>>
>>1) It sounds like you are doing stitching in microscopy or
>>    creating a mosaic for video or satelite images.
>>
>>                 Is that the case ?
>>
>>
>>2) The subject of your email says "Inter-Modality", but...
>>    if you are doning stitching, it is unlikely that your
>>    images are actually from multiple modalities...
>>
>>    Please clarify:
>>
>>          Are the images of the same modality or not ?
>>
>>
>>3) If you are doing stiching you can solve the registration
>>    of every pair of images by setting the FixedImageRegion
>>    to the expected overlap (5mm x 20mm) and by making sure
>>    that you assign to all the images, values of pixel spacing
>>    and image origin that correspond roughly to their correct
>>    location in space. That will simplify a lot the registration.
>>
>>
>>4) If you are doing this for Microscopy you may want to start
>>    with the following components:
>>
>>     - NormalizedCorrelationMetric
>>     - TranslationTransform
>>     - LinearInterpolator
>>     - RegularStepGradientDescentOptimizer
>>
>>
>>5) Make sure that you connect an Observer to the optimizer
>>    and you plot the values of the metric as the iterations
>>    proceed. That will guide in the process of fine tunning
>>    registration parameters. Keep in mind that you cannot
>>    trust an optimizer, you must supervise its progress and
>>    drive its parameters in order to obtain reasonable results.
>>    The use of command observers is illustrated in almost all
>>    the registration examples in
>>
>>               Insight/Examples/Registration
>>
>>
>>
>>
>>   Regards,
>>
>>
>>      Luis
>>
>>
>>
>>-------------------------
>>Wei-Lin Bee wrote:
>>
>>
>>>Hi,
>>>
>>>I have 64 spacial image files (8 X 8)to be registered. The size of each
>>
>>image is
>>
>>>20mm X 20mm which overlap partion between two images would be 5mm X 20mm.
>>
>>I
>>
>>>learned from itkSoftwareGuide that there seems no clear-cut rules in
>>
>>choosing
>>
>>>matric, however I failed to solve this problem for days. Any reply would
>>
>>be
>>
>>>appreciated.
>>>
>>>
>>>
>>>Thank you very much.
>>>
>>>
>>>Best,
>>>Tiger
>>>_______________________________________________
>>
>>
>>
>>
>>
> 
>