[Insight-users] Re: ITK Configuration

Luis Ibanez luis.ibanez@kitware.com
Thu May 20 16:37:33 EDT 2004


Hi Shekhar,


1) "Parameters setting" is the fundamental flaw of
    any attempt to compare methods in image processing.

    It seems that the community hasn't recognized
    yet the fact that parameters are integral part
    of a method. In some way the parameter values
    *are* part of the code.

    There are no standard configurations for
    registration. The best that you can do is

    a) Select a *very specific* image
       registration problem.

    b) Take a registration method configuration
       and fine tune its full set of parameters
       until you get the best results you can
       find.

    c) Take another registration method configuration
       and again fine tune its full set of parameters
       until you get the best results

    d) Compare the results of (b) and (c) in terms
       of computational time and stability, (e.g.
       sensibility to variations in the parameters).

    Note that the conclusions that you can draw at
    the end are of the type:

    Method A *with* this set of parameters behave
    better than Method B *with* this other set
    of parameter when registering *this* specific
    pair of images.

    Any gratuitous generalization of the type

        "Method A is better than Method B"

    will only be good for 'publishing a paper'...




2) Please read the chapter on Multi-Resolution
    image registration in the SoftwareGuide

    http://www.itk.org/ItkSoftwareGuide.pdf

    the number of levels you define in the MultiResolution
    registration method is not the number of pyramids to
    use. The number of pyramids is *always* two.  The value
    is used to define how many levels to use in those two
    pyramids.




3) Please read the chapter above....
    and try the associated examples in


    Insight/Examples/Registration/
              MultiResImageRegistration1.cxx
              MultiResImageRegistration2.cxx



4) NormalizedMutualInformation is available in ITK

http://www.itk.org/Insight/Doxygen/html/classitk_1_1NormalizedMutualInformationHistogramImageToImageMetric.html

    Please consult the Doxygen documentation whenever
    you are looking for specific methods in ITK. The
    best places to look are:

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

    and

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




5) Please read the Chapter on image registration in
    the SoftwareGuide. You will find there indications
    on how to set optimizer parameters.

    You may find also interesting to look at the
    course on Image Registration Techniques

    http://www.cs.rpi.edu/courses/spring04/imagereg/

    In particular to lectures 7,8,9 and 10.

    If you have questions about a *specific* optimizer
    please let us know.





Regards,



    Luis




-------------------------------------------
Dwivedi, Shekhar (Research, Non-ge) wrote:

> Hi luis,
> 
> there are few doubts , please help me.
> 
> 1- If we are comparing various optimizers or metrics or filters, then what is the standard configuration on whose basis we will change other parameters, I mean that there must be some parameters which should be fixed at a time and others should change. e.g. At a time either we use single image registration or multi image registration and then the paramteres are changed.
> 
> 2- The number of levels which we declare in the multi Image registration is the number of Image pyramids we are to consider or something else.IF we set the number of levels as 1 then will  it be same as single image registration (means no pyramid of images).
> 
> 3- If we set the number of levels as 2 then only initialization of the pyramid filter will mean that registration will take care of generating the pyramids(gaussians) or we have to specify the parameters of the same.
> 
> 4- Is Normalised Mutual Information implemented in ITK.
> 
> 5- As for each optimizer , every parameter is not set for registration unlike as stated in the test cases, can I found any documentation which helps me in understanding the parameters effecting in specific Image registration.
> 
> yours,
> shekhar.
> 






More information about the Insight-users mailing list