[Insight-users] Rigid3DTransform

[Luis Ibanez]

Hi SriValli,

Ridig3DTransform is rather an abstract class.
It simply factorizes the functionalities of
several classes that parameterize the transform
space in differen ways.


Please use any of its derived classes:

http://www.itk.org/Insight/Doxygen/html/classitk_1_1Rigid3DTransform.html


- Euler3DTransform
- QuaternionRigidTransform
- VersorRigid3DTransform
- VersorTransform



Probably the best adapted for rigid 3D registration
is VersorRigid3DTransform.
http://www.itk.org/Insight/Doxygen/html/classitk_1_1VersorRigid3DTransform.html

Note that this transform *must* be used in conjunction with
the optimizer,
http://www.itk.org/Insight/Doxygen/html/classitk_1_1VersorRigid3DTransformOptimizer.html
because Versors *do not* form a vector space

Along the same lines, the transform
http://www.itk.org/Insight/Doxygen/html/classitk_1_1QuaternionRigidTransform.html
must be used along with the optimizer
http://www.itk.org/Insight/Doxygen/html/classitk_1_1QuaternionRigidTransformGradientDescentOptimizer.html


The critical parameters in a 3D registration process are:


- Learning rate (if you use GradientDescent optimizer)
- StepLength    (if you use RegularStepGradientDescent optimizer)


- Translation scale : which defines the ratio between rotations and
                       translation during optimization.



Regards,


     Luis



------------------------
valli gummadi wrote:

>  Hi Mr.Luis,
>            I tried registration using CenteredRigid2dTransform class and achieved results. I Used centeredAffineTransform for 3d Registration.
> My doubt is Can I use the Rigid3dTransform class for 3D Registration.
> Because, When i used this class , I am gettign runtime error at 
> transform->GetParameters().
> 
> Code Snippet:
> 
> typedef itk::Rigid3DTransform<double> TransformType;
> TransformType::Pointer	transform	= TransformType::New();
> short NoofParams;
> NoofParams = transform->GetNumberOfParameters();itk::Array<double> Params( NoofParams );
> Params.Fill( 0 );
> transform->SetIdentity();
> Params = transform->GetParameters();
> 
> Another problem is:
> In ITK, Many parameters are available for registration to tune to get teh desired output. But, I am always get confusion which parameters to tune to get better output based on the image properties. In ITKSoftware Guide.pdf,I found a statement "Both images have unit-spacing and are shown in
> Figure 8.17. In order to accelerate convergence it is convenient to use a larger step length".
> This statement is quite helpful to set max step length.
> Can you please explain how to set remaining parameters based on the image.
> This will be more help ful for me to tune my registration so that i can register any set of images just by changing params.
> 
> Thank you,
> Regards,
> SriValli.
>