[Insight-users] optimizer scales with rectangular 2D images

Thu Jan 19 14:12:16 EST 2012

Hi, 

I want to find the 3 parameters of Euler 2D Transform that register my 2 spect images.

I thought I understood the use of optimizer scales correctly. But definitively not. 

If I initialize my transform with 0 for all my 3 parameters:
  typedef RegistrationType::ParametersType ParametersType;
  ParametersType initialParameters(transformParametersTotal);
  initialParameters.Fill(0.0);
  registration_->SetInitialTransformParameters(initialParameters);

, and define scales as follow:
  typedef OptimizerType::ScalesType ScalesType;
  ScalesType parametersScales(3);
  parametersScales.Fill( 1.0 );
  for (int j = 1; j < 3; j++ ){
    parametersScales[j] = 1.0 / 1000;
  }
  optimizer_->SetScales(parametersScales);

How come my first iteration test for such parameters:
-------------------------------------
MultiResolution Level : 0
0 = -0.372419 : [1.7683, -0.0955618, -0.929501]
1 = -0.0383217 : [1.92583, -2.08874, -0.978926]
2 = -0.0362538 : [1.69325, -1.10031, -2.21542]
3 = -0.0377709 : [1.56321, -1.28384, -3.79953]
4 = -0.0419196 : [1.66387, -2.53705, -3.55923]
5 = -0.0393448 : [1.52317, -2.40928, -4.56544]

I have many questions:
A) My images are 256 x 1024 pixels, with pixel size of 2.6800001 mm/pixel. Do I have to take the fact that images are rectangular into account? Is that why the first parameter that is the rotation seem to change more often than translation parameters?

B) In one of your previous mails: http://itk-insight-users.2283740.n2.nabble.com/Confused-abour-Optimizer-Scales-td4010857.html
it says that if you don't want one of the parameters to change (for example, center of rotation in CenteredRigid2DTransform to have to set to very small number
"The second and third values correspond to the center of rotation, 
     but you don't want to change this center.  My suggestion will be 
     to put in here, very small numbers. For example  1e-6. "

Is it a small number or a very large one?

I am really stuck with all of this. Can somebody help me?

Regards,

Isabelle

________________________________
 De : Renaud Isabelle <renauisa at yahoo.fr>
À : "insight-users at itk.org" <insight-users at itk.org> 
Envoyé le : Mercredi 18 Janvier 2012 23h34
Objet : [Insight-users] initialization transform parameters not taken into account

Hi all,

I am trying to register 2D images (spect modality) witk itk Mattes mutual information. Visually, offsets between both images are very small (translation of ~ 5 mms). I am initializing transform parameters to 0 for the 3 parameters of euler 2D transform. However, after registration I got a rotation around 90 degres. Plus, when I am observing optimizer iterations, first iteration is already 1.6 radian for rotation parameter, even if my initial rotation parameter was 0. 

Here is the observer:

MultiResolution Level : 0
0 = -0.372419 : [1.7683, -0.0955618, -0.929501]
1 = -0.0383217 : [1.92583, -2.08874, -0.978926]
2 = -0.0362538 : [1.69325, -1.10031, -2.21542]
3 = -0.0377709 : [1.56321, -1.28384, -3.79953]
4 = -0.0419196 : [1.66387, -2.53705, -3.55923]
141 = -0.042436 : [1.51798, -17.6144, -6.53557]
142 = -0.0423536 : [1.52173, -17.635, -6.52587]
143 = -0.042424 : [1.51635, -17.6507, -6.54188]

 Iterations    = 145
 Metric value  = -0.0424182
 Stop Condition  = RegularStepGradientDescentOptimizer: Step too small after 144
 iterations. Current step (0.0184467)
 is less than minimum step (0.02).
Result =
 Translation X = 17.6507 mm
 Translation Y = 6.54188 mm
 Rotation X = -86.8805 degres

Here is part of the print out of registration:
  CurrentLevel: 1
  InitialTransformParameters: [0, 0, 0]
  InitialTransformParametersOfNextLevel: [0, 0, 0]
  LastTransformParameters: [1.51635, -17.6507, -6.54188]
  FixedImageRegion: ImageRegion (103F17EC)
  Dimension: 2
  Index: [0, 0]
  Size: [256, 1024]

  FixedImageRegion at level 0: ImageRegion (04DA4C98)
  Dimension: 2
  Index: [0, 0]
  Size: [128, 512]

  FixedImageRegion at level 1: ImageRegion (04DA4CAC)
  Dimension: 2
  Index: [0, 0]
  Size: [256, 1024]

It's clear that I am missing something. Please help. Here is my code:

  ImageType::RegionType fixedImageRegion = fixedImage->GetBufferedRegion();
  ImageType::RegionType movingImageRegion = movingImage->GetBufferedRegion();   

  typedef itk::Euler2DTransform< double > TransformType;  
  TransformType::Pointer transform = TransformType::New();
  registration_->SetTransform(transform);

  // Initialize transform parameters
  int transformParametersTotal =
 registration_->GetTransform()->GetNumberOfParameters();
  typedef RegistrationType::ParametersType ParametersType;
  ParametersType initialParameters(transformParametersTotal);
  initialParameters.Fill(0.0);
  registration_->SetInitialTransformParameters(initialParameters);  

  metric_->SetNumberOfHistogramBins(50);
  metric_->UseAllPixelsOn();

  registration_->SetFixedImage(fixedImage);
  registration_->SetMovingImage(movingImage);    
  registration_->SetFixedImageRegion(fixedImageRegion);

  typedef OptimizerType::ScalesType ScalesType;
  int transformParametersTotal = registration_->GetTransform()->GetNumberOfParameters();
  ScalesType parametersScales(transformParametersTotal);
  parametersScales.Fill( 1.0 );

  for (int j = 1; j < transformParametersTotal; j++ ){
    parametersScales[j] = 1.0 / 1000;
  }        

  optimizer_->SetScales(parametersScales);

  // The initial step length is defined with SetMaximumStepLength()
  optimizer_->SetMaximumStepLength(2.0);  
  optimizer_->SetMinimumStepLength(0.02);

  // Define number of iterations per multi-resolution level
  optimizer_->SetNumberOfIterations( 200);

  optimizer_->SetRelaxationFactor(0.8);     

  OptimizerIterationUpdate::Pointer
 observer = OptimizerIterationUpdate::New();
  optimizer_->AddObserver( itk::IterationEvent(), observer );

   typedef RegistrationInterfaceCommand<RegistrationType> CommandType;
   CommandType::Pointer command = CommandType::New();
   registration_->AddObserver( itk::IterationEvent(), command );

    registration_->Update(); 
    // Output parameters
    ParametersType finalParameters = registration_->GetLastTransformParameters();
     result[0] = ;
      result[1] = 
      result[2] = ;                   

    // Print out results
    std::cout << "Result = " << std::endl;
    std::cout
 << " Translation X = " << - finalParameters[1]  << std::endl;
    std::cout << " Translation Y = " <<- finalParameters[2];<< std::endl;
     std::cout << " Rotation X = " << -finalParameters[0] * RAD_TO_DEG<< std::endl;  

Isabelle Renaud

_____________________________________
Powered by www.kitware.com

Visit other Kitware open-source projects at
http://www.kitware.com/opensource/opensource.html

Kitware offers ITK Training Courses, for more information visit:
http://www.kitware.com/products/protraining.html

Please keep messages on-topic and check the ITK FAQ at:
http://www.itk.org/Wiki/ITK_FAQ

Follow this link to subscribe/unsubscribe:
http://www.itk.org/mailman/listinfo/insight-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20120119/744d8ab6/attachment.htm>