Normalize Image#

Synopsis#

Normalize an image.

Results#

../../../../_images/Yinyang25.png — Input Image#

../../../../_images/NormalizeImage.png — Input And Output Image With Data#

Code#

C++#

#include "itkImage.h"
#include "itkImageFileReader.h"
#include "itkNormalizeImageFilter.h"
#include "itkStatisticsImageFilter.h"

#ifdef ENABLE_QUICKVIEW
#  include "QuickView.h"
#endif

#include <iomanip>

int
main(int argc, char * argv[])
{
  if (argc < 2)
  {
    std::cerr << "Usage: " << argv[0] << " filename" << std::endl;
    return EXIT_FAILURE;
  }

  using FloatImageType = itk::Image<double, 2>;

  const auto input = itk::ReadImage<FloatImageType>(argv[1]);

  using NormalizeFilterType = itk::NormalizeImageFilter<FloatImageType, FloatImageType>;
  auto normalizeFilter = NormalizeFilterType::New();
  normalizeFilter->SetInput(input);

  using StatisticsFilterType = itk::StatisticsImageFilter<FloatImageType>;
  auto statistics1 = StatisticsFilterType::New();
  statistics1->SetInput(input);

  auto statistics2 = StatisticsFilterType::New();
  statistics2->SetInput(normalizeFilter->GetOutput());

#ifdef ENABLE_QUICKVIEW
  QuickView viewer;

  std::stringstream desc1;
  statistics1->Update();
  desc1 << itksys::SystemTools::GetFilenameName(argv[1]) << "\nMean: " << statistics1->GetMean()
        << " Variance: " << statistics1->GetVariance();
  viewer.AddImage(input.GetPointer(), true, desc1.str());

  std::stringstream desc2;
  statistics2->Update();
  desc2 << "Normalize"
        << "\nMean: " << std::fixed << std::setprecision(2) << statistics2->GetMean()
        << " Variance: " << statistics2->GetVariance();
  viewer.AddImage(normalizeFilter->GetOutput(), true, desc2.str());

  viewer.Visualize();
#endif
  return EXIT_SUCCESS;
}

Classes demonstrated#

template<typename TInputImage, typename TOutputImage> class NormalizeImageFilter : public itk::ImageToImageFilter<TInputImage, TOutputImage>

Normalize an image by setting its mean to zero and variance to one.

NormalizeImageFilter shifts and scales an image so that the pixels in the image have a zero mean and unit variance. This filter uses StatisticsImageFilter to compute the mean and variance of the input and then applies ShiftScaleImageFilter to shift and scale the pixels.

NB: since this filter normalizes the data such that the mean is at 0, and $-\sigma$ to $+\sigma$ is mapped to -1.0 to 1.0, output image integral types will produce an image that DOES NOT HAVE a unit variance due to 68% of the intensity values being mapped to the real number range of -1.0 to 1.0 and then cast to the output integral value.

See

NormalizeToConstantImageFilter

ITK Sphinx Examples:

All ITK Sphinx Examples

Normalize Image

See itk::NormalizeImageFilter for additional documentation.