[Insight-users] registration using rotation + translation

[Luis Ibanez]

Hi Kevin

On Tue, Nov 30, 2010 at 1:23 PM, Kevin Neff <kevin.l.neff at gmail.com> wrote:
>
> Thanks, Dan.  I used the example ImageRegistration8.cxx with my data.
>
> First, it's terribly slow.  Is there anything I can do to speed it up?


1)  How long it took ?

2)  How many cores do you have ?

3) What version of ITK are you using ?

4) Make sure that you compile it for Release mode.
     (that easily makes a difference of 10x with respect to Debug mode).


> my limited experience, applying a threshold first and skipping rotation of
> voxels below the threshold helped.  Or could I pre-compute isotropic voxels
> and skip the interpolation that follows image rotation and just interpolate
> the final solution?  Any suggestion is welcome!
>

You could also use a Mask,
See example:

ITK/Examples/Registration/ImageRegistration12.cxx


> Second, after 200 iterations,  the output does not appear correct.  I ran it
> with data that had the dimensions 13888x560x48 and 1388x560x53 but the
> output had the dimensions 13888x560x48.  The volumes were of the shell of a
> large object, rotated 90 deg, so I'm sure the dimension of the output should
> have been much larger---something like 1388x1000x100.

a) 90 degrees is beyond the capture radio of any registration method.
    You must initialize the Transform to account for most of that rotation.
    In practice you shouldn't expect the optimization process to correct
    for more than 20 to 30 degrees of rotation.

b) Is that rotation along the long axis of your image ? or perpendicular to it ?
    A screen shot of your image will be very helpful...

c) You should carefully control the parameter scales to make sure that
    you account for the proportion of radians units (used for rotation) to
    millimeters used for translation.

d) What is the pixel spacing of your images ?


>
> KLN
>
>
>
>
>
> On Sun, Nov 21, 2010 at 7:55 AM, Dan Mueller <dan.muel at gmail.com> wrote:
>>
>> Hi Kevin,
>>
>> I highly recommending the entire ITK Software Guide:
>>    http://www.itk.org/ItkSoftwareGuide.pdf
>>
>> For your specific question, see section 8.8.
>>
>> In short:
>>
>> IdentityTransform: identity
>> TranslationTransform: n-D translation
>> ScaleTransform: n-D scaling
>> CenteredRigid2DTransform: 2-D rotation + translation
>> Similarity2DTransform: 2-D scaling + rotation + translation
>> VersorTransform: 3-D rotation
>> VersorRigid3DTransform: 3-D rotation + translation
>> Similarity3DTransform: 3-D scaling + rotation + translation
>> AffineTransform: n-D scaling + rotation + translation + shear
>>
>> HTH
>>
>> Cheers, Dan
>>
>> On 21 November 2010 13:49, Kevin Neff <kevin.l.neff at gmail.com> wrote:
>> >
>> > I need to register images using rotation and translation.  Is that
>> > possible
>> > with theITK classes, or do I have to write my own?
>> >
>> > I'm basing this question on a translation-only example and the list of
>> > classes derived from AffineTransform
>> >
>> > http://www.itk.org/Wiki/ITK_Image_Registration
>
>
> _____________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Kitware offers ITK Training Courses, for more information visit:
> http://www.kitware.com/products/protraining.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
>
>