[Insight-users] Sparse image for building distance field

[Sergiy Tkachuk]

Hello all,

I want to build distance field for the object.
My object is very sparsed and I want to build distance field for it and not
to allocate memory for every voxel in the grid.
Number of voxel in 3d grid is about 10^10 (more precisely: 2000x2000x2000)
Number of voxel in object itself is about 10^6.

I tried to play with itk::SignedMaurerDistanceMapImageFilter creating image
5x5x5 in sandbox:


    const unsigned int Dimension = 3;
    typedef char InputPixelType;
    typedef float OutputPixelType;

    typedef itk::Image<InputPixelType, Dimension> InputImageType;
    typedef itk::Image<OutputPixelType, Dimension> OutputImageType;

    InputImageType::Pointer image = InputImageType::New();

    InputImageType::SizeType size;
    size[0] = 5;
    size[1] = 5;
    size[2] = 5;

    InputImageType::IndexType start;
    start.Fill(0);

    InputImageType::RegionType region;
    region.SetIndex(start);
    region.SetSize(size);

    image->SetRegions(region);
    image->Allocate();

    image->FillBuffer(0);

    InputImageType::IndexType index;

    // initialize
    InputImageType::IndexType gapIndex;
    gapIndex[0] = gapIndex[1] = gapIndex[2] = 2;
    for (index[0] = 0; index[0] < size[0]; ++index[0])
        for (index[1] = 0; index[1] < size[1]; ++index[1])
            for (index[2] = 0; index[2] < size[2]; ++index[2]) {
                if (index == gapIndex) continue;
                image->SetPixel(index, 1);
            }

    // Define the distance map filter and apply it to the image
    typedef itk::SignedMaurerDistanceMapImageFilter<InputImageType,
OutputImageType> DistanceFilterType;
    DistanceFilterType::Pointer distanceFilter = DistanceFilterType::New();
    distanceFilter->SetInput(image);
    distanceFilter->SetBackgroundValue(0);
    distanceFilter->SetSquaredDistance(false);
    distanceFilter->Update();

    // Output distance field values
    for (int x = 1; x < size[0]-1; ++x)
    {
        for (int y = 1; y < size[1]-1; ++y)
        {
            for (int z = 1; z < size[2]-1; ++z)
            {
                InputImageType::IndexType ind;
                ind[0] = x;
                ind[1] = y;
                ind[2] = z;

                float value = distanceFilter->GetOutput()->GetPixel(ind);
                std::cout << x << " " << y << " " << z << " -> " << value
<< std::endl;
            }
        }
    }


The code works fine.
But it requires memory for every voxel of grid, and this is not acceptable
for 10^10 voxels.

Is it possible to use some kind of sparse structure to build distance field?

I read about itk::SparseImage, but it seems this structure is sparse
concerning value only, but indices are allocated for the whole grid. Am I
undestood correctly? If no, I will be glad if somebody point out where I am
mistaken. Thanks.

If itk::SparseImage is not sutable in my case, is it possible to implement
custom Image format for input to itk::SignedMaurerDistanceMapImageFilter?
If so, are the any useful guidelines to achieve this?
For now it is not clear for me how to correctly derive my Image type from
DataObject and handle regions - I don't need full power of ITK filter
pipeline, just to call itk::SignedMaurerDistanceMapImageFilter - possibly
this helps me to simplify custom image implementation.

Any ideas could be helpful for me!

Thanks in advance,
Sergiy
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20130322/4515936d/attachment.htm>