[Insight-users] Registration and Spacing : Bug

[David Holmes]

Luis-

I apologize.  I may be slow but eventually I catch on.
 I just realized that I had a major blunder that
better explains why I am having trouble.  So I moved
to the translation transform per your suggestion and
started fooling around with it.  It was still failing,
but I realized that I am just dumb.  I was using
multi-resolution because it is so much quicker.  Given
that the spacing of my data is [1,1,1] and
registration is in real-world coordinates, then my
pyramid isn't going to work because of the spacing on
the lower resloution levels.  If I get rid of the
pyramid, the translation works.  So, all of this time,
I have been trying to fix the lowest-resolution pass
of the multi-res registration, instead of looking at
the full resolution.  I'm going to take a look at the
versor one now and see if that is similar.

Thanks

david


--- Luis Ibanez <luis . ibanez at kitware . com> wrote:
> 
> Hi David,
> 
> Another possible suspect is the computation of
> Jacobians
> in the transforms. However, the translation
> transform
> used by Carolyn and Michael have a quite simple
> Jacobian,
> which leaves few room for error.
> 
> In the case you mention, the VersorRigid3DTransform
> is used,
> this transform has a rather complex Jacobian, which
> make it
> a better candidate for failure.
> 
> It is not clear at this point if both effects of
> increasing
> cost-function are originated from a single error.
> 
> Would you observe the same miss-behavior of the
> optimizer
> if you replace the VersorRigid3DTransform with a
> simple
> Translation  ?
> 
> 
> 
> Regards,
> 
> 
> 
>     Luis
> 
> 
> 
> ---------------------
> David Holmes wrote:
> > Luis-
> > 
> > I am not sure that it is simply an issue with
> non-unit
> > spacing.  I have had this problem for some time
> (refer
> > to
> >
>
http://www . itk . org/pipermail/insight-users/2003-February/002637 . html).
> >  To confirm the problem again, I used the
> > ImageRegistration8.cxx example with a very simple
> > dataset and show the same results (i.e. the cost
> value
> >  increases).  In my case, the spacing is [1,1,1]
> for
> > both fixed and moving.  I've fooled with
> parameters. 
> > That is not the problem.  I've run both simple and
> > multi-resolution registration.  That is not the
> > problem.
> > 
> > thanks
> > 
> > david
> > 
> > 
> > --- Luis Ibanez <luis . ibanez at kitware . com> wrote:
> > 
> >>Hi Michael, Carolyn
> >>
> >>We were finally able to duplicate the problem you
> >>both
> >>reported concerning the registration of image data
> >>with
> >>non-unit spacing.
> >>
> >>
> > 
> >
>
http://www . itk . org/pipermail/insight-users/2003-July/004215 . html
> > 
> >
>
http://www . itk . org/pipermail/insight-users/2003-July/004314 . html
> > 
> >>Using a simple registration case with translation
> >>transform,
> >>we found that a very small spacing causes the
> >>optimizer to
> >>become erratic.
> >>
> >>We are tracking this problem in order to find a
> fix.
> >>
> >>Our guess is that the spacing is not correctly
> >>considered
> >>when computing the metric derivatives in the
> >>registration
> >>method. A bug report has been filed in GNATS.
> >>
> >>
> >>  Thanks for pointing this out.
> >>
> >>
> >>     Luis
> >>
> >>
> >>
> >>
> >>_______________________________________________
> >>Insight-users mailing list
> >>Insight-users at itk . org
> >>http://www . itk . org/mailman/listinfo/insight-users
> > 
> > 
> > 
> > __________________________________
> > Do you Yahoo!?
> > Yahoo! SiteBuilder - Free, easy-to-use web site
> design software
> > http://sitebuilder . yahoo . com
> > 
> 
> 
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk . org
> http://www . itk . org/mailman/listinfo/insight-users


__________________________________
Do you Yahoo!?
Yahoo! SiteBuilder - Free, easy-to-use web site design software
http://sitebuilder . yahoo . com