[Insight-users] Versor Optimizer

[Luis Ibanez]

Hi Ravi,

You are right, the exception is the result of mapping
the moving image outside of the fixed image domain.

The reason why the transform is going out of range
is that this optimizer is incompatible with the
transform you are using.

The VersorOptimizer is intended to be used with
the VersorTransform, which is a pure translation
transform.

We are in lack of an optimizer specific for
VersorRigid3D transform. This is nedeed because the
parameter space of the VersorRigid3D is 6 dimensional,
but it is not a full vector space. 3 of its dimensions
are a versor space and the other three are vector ones.

This means that a gradient-descent algorithm has to
apply updates in different ways for the two sets of
3 dimensions.

In the meantime,
you may want to use the QuaternionRigid transform
and its associated optimizer:

http://www . itk . org/Insight/Doxygen/html/classitk_1_1QuaternionRigidTransform . html
http://www . itk . org/Insight/Doxygen/html/classitk_1_1QuaternionRigidTransformGradientDescentOptimizer . html


Please let us know if your encounter any problem,


Thanks


    Luis


----------------------
ravi sharma wrote:
> Hi,
> 
>  I am trying 3D Rigid Body Registration using Versor
> Rigid3DTransform and VersorOptimizer.As a first step I
> have taken the source and target image to be the same
> image.I am using MattesMutualInformation Metric.The
> problem is that I try with initial parameters of
> VersorTransform as 0.01 i +0.01 j+0.01 k so that
> rotaion angle is almost zero(translation parameters
> are all 0.0).I get the following exception:
> 
> ExceptionObject caught !
> 
> itk::ExceptionObject (0x21f988)
> Location: "Unknown" 
> File:
> /opt/tools/itk/Insight_Nightly/Insight_1_2/Insight/Code/Algorithms/itkMattesMutualInformationImageToImageMetric.txx
> Line: 546
> Description: itk::ERROR:
> MattesMutualInformationImageToImageMetric(0x21c6f0):
> Too many samples map outside moving image buffer: 6602
> / 75000
>   I think the problem could be that my initial
> parameters lead the image to map out of each other or
> the step sizes of the optimizer are too large.
> 
>  What could be safe minimum and maximum step sizes???
> 
> Kind Regards,
> Ravi Sharma.
> 
> 
>