itk中的花式数据切割（四）

之前一直是自定义的图像数据切切切，其实itk中已经给我们提供了一个专门来提取切片的类：itkExtractImageFliter。今天深入解析下官方的思路是什么样的，有哪些是我们可以借鉴的。

ImageType::RegionType inputRegion =	input_data->GetLargestPossibleRegion();

ImageType::SizeType size = inputRegion.GetSize();
size[2] = 0;// 提取垂直于Z轴的切片（XYZ轴随意）

// 设置切片在Z轴的位置及切片大小与起始索引
ImageType::IndexType start = inputRegion.GetIndex();
const unsigned int sliceNumber = 2;
start[2] = sliceNumber;

ImageType::RegionType desiredRegion;
//无聊的话可以试试自己改变size和start，温馨提示：中间有坑，新手司机请注意路况
desiredRegion.SetSize(  size  );
desiredRegion.SetIndex( start );

// 创建提取图像滤波器
typedef itk::ExtractImageFilter< ImageType, ImageType > FilterType;
FilterType::Pointer filter = FilterType::New();
filter->SetInput( input_data);
// 设置提取切片区域
filter->SetExtractionRegion( desiredRegion );

output_data = filter->GetOutput()

这里要注意一下（敲黑板，划重点）：itkExtractImageFliter提取出来的切片是二维的，不是在三维空间中。

切片的 physical dimensions:185*185*0

原始数据的 physical dimensions:185*185*90.5

区别就在于Z轴变成了0，第三个维度消失了~成功完成了三维空间到二维空间的转换。（这儿出现了一个问题：怎么转换回去呢？）

接下来把itkExtractImageFliter这个类打开看下官方是怎么实现的。没多少代码，txx中总共282行，4个函数，三个有用的：

1.SetExtractionRegion（设置ROI范围）

template 
void 
ExtractImageFilter
::SetExtractionRegion(InputImageRegionType extractRegion)
{
  m_ExtractionRegion = extractRegion;

  unsigned int nonzeroSizeCount = 0;
  InputImageSizeType inputSize = extractRegion.GetSize();
  OutputImageSizeType outputSize;
  OutputImageIndexType outputIndex;

  /**
   * check to see if the number of non-zero entries in the extraction region
   * matches the number of dimensions in the output image.  
   */
  for (unsigned int i = 0; i < InputImageDimension; ++i)
    {
    if (inputSize[i])//这里就是降维的秘密，主动忽略一个维度
      { 
      outputSize[nonzeroSizeCount] = inputSize[i];
      outputIndex[nonzeroSizeCount] = extractRegion.GetIndex()[i];
      nonzeroSizeCount++;
      }
    }
    
  if (nonzeroSizeCount != OutputImageDimension)
    {
    itkExceptionMacro("Extraction Region not consistent with output image");
    }

  m_OutputImageRegion.SetSize(outputSize);
  m_OutputImageRegion.SetIndex(outputIndex);
  this->Modified();
}

2.GenerateOutputInformation（生成输出数据的信息）

（代码太长，只贴一部分有用的）

    unsigned int i;
    const typename InputImageType::SpacingType& 
      inputSpacing = inputPtr->GetSpacing();
    const typename InputImageType::DirectionType&
      inputDirection = inputPtr->GetDirection();
    const typename InputImageType::PointType&
      inputOrigin = inputPtr->GetOrigin();

    typename OutputImageType::SpacingType outputSpacing;
    typename OutputImageType::DirectionType outputDirection;
    typename OutputImageType::PointType outputOrigin;

//这个贴心的if~ 作者考虑到有人故意捣乱，想把3维数据转换为4维数据。帅
    if ( static_cast(OutputImageDimension) > 
         static_cast(InputImageDimension )    )
      {
      // copy the input to the output and fill the rest of the
      // output with zeros.
      for (i=0; i < InputImageDimension; ++i)
        {
        outputSpacing[i] = inputSpacing[i];
        outputOrigin[i] = inputOrigin[i];
        for (unsigned int dim = 0; dim < InputImageDimension; ++dim)
          {
          outputDirection[i][dim] = inputDirection[i][dim];
          }
        }
      for (; i < OutputImageDimension; ++i)
        {
        outputSpacing[i] = 1.0;
        outputOrigin[i] = 0.0;
        for (unsigned int dim = 0; dim < InputImageDimension; ++dim)
          {
          outputDirection[i][dim] = 0.0;
          }
        outputDirection[i][i] = 1.0;
        }
      }
    else//提取切片的代码主要集中在这里
      {
      // copy the non-collapsed part of the input spacing and origing to the output
      outputDirection.SetIdentity();
      int nonZeroCount = 0;
      for (i=0; i < InputImageDimension; ++i)
        {
        if (m_ExtractionRegion.GetSize()[i])
          {
          outputSpacing[nonZeroCount] = inputSpacing[i];
          outputOrigin[nonZeroCount] = inputOrigin[i];
          int nonZeroCount2 = 0;
          for (unsigned int dim = 0; dim < InputImageDimension; ++dim)
            {
            if (m_ExtractionRegion.GetSize()[dim])
              {
              outputDirection[nonZeroCount][nonZeroCount2] =
                inputDirection[nonZeroCount][dim];
              ++nonZeroCount2;
              }
            }
          nonZeroCount++;
          }
        }
      }
    // This is a temporary fix to get over the problems with using OrientedImages to
    // a non-degenerate extracted image to be created.  It still needs to be determined
    // how to compute the correct outputDirection from all possible input directions.
    if (vnl_determinant(outputDirection.GetVnlMatrix()) == 0.0)
      {
      outputDirection.SetIdentity();
      }

    // set the spacing and origin
    outputPtr->SetSpacing( outputSpacing );
    outputPtr->SetDirection( outputDirection );
    outputPtr->SetOrigin( outputOrigin );
    outputPtr->SetNumberOfComponentsPerPixel(inputPtr->GetNumberOfComponentsPerPixel() );

3.ThreadedGenerateData（对应数据填灰度值）

（涉及到多线程部分暂时不看）

  // walk the output region, and sample the input image
  while( !outIt.IsAtEnd() )//遍历
    {
    // copy the input pixel to the output 对，就是这里
    outIt.Set( static_cast(inIt.Get()));
    ++outIt; 
    ++inIt; 
    progress.CompletedPixel();
    }

看来图像处理的基础部分，不论是itk官方还是初学者，写起来都差不多么。都是设置范围，遍历，循环，对应像素点，没了~

参考文献：

1. https://itk.org/Wiki/ITK/Examples/ImageProcessing/ExtractImageFilter

2.《三维图像编程实验》