[Insight-users] Re: scale factor in registration methods

[Silvester Czanner]

> Luis,
>
> thank you for your explanation. I tried to "play" with the example, 
> you recommended me, but  I was not successful. :-(
> I registered the ITK images:
>
> [czanner@gabi test]$ ~/CVSROOT/Insight_bin/bin/ImageRegistration7 
> BrainProtonDensitySliceBorder20.png 
> BrainProtonDensitySliceR10X13Y17.png a.mha
>
> here is my result:
> Result =
> Scale         = -6.87933e-08
> Angle         = 1.42343e-08
> Center X      = 110.281
> Center Y      = 128.664
> Translation X = 122.5
> Translation Y = 144.5
> Iterations    = 423
> Metric value  = 21561.4
>
> It is far away from the reality... . Could you, please recommend me 
> some setup for this method, I use similar parameters as it is in other 
> examples .
>
> Is this ( I mean the Similarity2DTransform ) the only method, where it 
> is possible to register images with different sizes ?
>
> Thank you,
> Silvester
>
> Luis Ibanez wrote:
>
>> Hi Silvester,
>>
>> Setting the optimization-scale factors for Scaling
>> is even harder than for rotations.  The reason is
>> that scales should in principle be treated in a
>> logarithmic framework.
>>
>> The following example uses the Similarity Transform,
>> which does Rotation, Translation and Scaling:
>>
>> Insight/Examples/Registration/ImageRegistration7.cxx
>>
>>
>> You may want to keep smaller step sizes when Scale
>> Factors are present in the transform, because the
>> non-linearity of the scale makes the optimization
>> quite unstable.
>>
>> This involves a laborious work of parameter
>> tunning... so... don't expect it to work at the
>> first try.
>>
>>
>> Please let us know if you find any problem.
>>
>>    Thanks
>>
>>      Luis
>>
>> -----------------------------
>> Silvester Czanner wrote:
>>
>>> Hi Luis,
>>>
>>> I am trying to use the registrations methods from ITK. The 
>>> assumption is, that there will be some small rotation, shifting and 
>>> also scaling. I received quite good results, when the registration 
>>> images have the same size, but there is a problem with scaling 
>>> factor. When I look to the examples there are use only images with 
>>> shifting and rotation. Where can I find how to setup or use the 
>>> scaling factor in registration?
>>>
>>> Thank you,
>>>
>>> Silvester
>>>
>>
>>