[Insight-users] Re: MultiResMIRegistration

[Christos Panagiotou]

Dear Luis

thanks again for a very detailed post :)
some questions about your answers!

Luis Ibanez wrote:

>
> Hi Christos,
>
> Isn't it fun to find good
> parameters for registration ?     :-)
>
>
> ----
>
> About your questions:
>
> 0)  As you pointed out, the factor is
>     a "suggestion". The relative scaling
>     between the translations and rotation
>     also allows you to control how fast to
>     advance in rotation with respect to
>     how much to advance in translations.
>
>     Depending on the predominant type of
>     mis-registration present in your images,
>     different range of factors may me more
>     convenient.
>
>     The suggestion is simply a reasonable
>     parameter from which you can start the
>     fascinating process of fine tunning the
>     values for your specific types of images.
>
>
> A)  There is not an inherent problem with the
>     MIMApplicationBase.txx way of computing the
>     scaling parameters. This values is a reasonable
>     first guess. You may use double of it, or half
>     of it and still be able to tune other registration
>     parameters in order to make the optimizer converge.
>
>
> B)  I don't see why you suggest to make the scaling
>     equal to  "131748". it seems that you forget the
>     sqrt() from your own computation, where you found
>     that the scaling should be around 363.

-----------------------------------
well 131748 is the value before the sqrt
MIMAppBase.txx takes the input scale factor (in this case 320) and then 
it computes the sqrt
of the input

so its 1.0 / sqrt(320)
its like sqrt(sqrt(11748)) and it is a bit different than 1.0 / 363
thats why i ve asked...
----------------------------------

>
>
>
> E)  Mutual Information may be quite counter-intuitive when
>     you apply it to trivial data, like the binary images
>     that you mention.  What you are asking to minimize is
>     the sparseness of the joint histogram. The optimizer
>     may end up selecting any transform parameters that lead
>     to such minimization. That includes to put all the white
>     of the square over all the black of the rectangle and
>     viceversa, since in this case you get a diagonal histogram.
>     And, although most people would consider that solution a
>     dramatic failure of the registration, well, that's just
>     a valid result for minimzing Mutual Information.


--------------
well about histogram... i mostly use the viola method
i ve tried to run the mattes implementation and i get an exception
 Requested region is (at least partially) outside the largest possible 
region. (how can i overcome this or what does this mean?
it seams to be invariant in the selection of histogram bins or spatial 
sample number (i use ITK 1.6)?)

well the optimizer should idealy find the "global" minimum of negative 
MI so the all white of the square over all the black
of the rectangle globaly minimizes only when there is exact registration 
in this case (am i wrong?)
h(a)+h(b)-h(a,b) (viola implementation)

what i get as a result in my case is a cube (moving image) affinly 
transformed to a paralelogram (fixed image)
the resampled moving image (a sheared paralelogram) fully overlaps the 
fixed however its ... sheared!
weeeell... as you said its probably optimization... i ve tried loads of 
attempts with varying learning rates and iterations and multi resolution 
levels but i can get the exact
result... and i thought it would be a simple test the cube/para case!
maybe mattes implementation (after getting rid of the exception) would 
help me to opimize it due to the smoothness of the metric.

i ll try to post some histograms when i generate them

really thanks for your reply
christos