[Insight-developers] An idea for new filter to get around TileImageFilter memory requirements

[Kent Williams]

I recently replaced some inline pixel copying code in our 4D 
OrientImageFilter with the TileImageFilter. The great strength of this 
approach is that the resulting code is very clear and concise. Basically 
the filter works this way:

Assume a 4D image as input, which represents a series of volumes.
for each i < # of volumes
  current imageVolume = ( ExtractImageFilter(input) | OrientImageFilter )
  tileImageFilter->SetNthInput(i, current imageVolume)
end
tileImageFilter->Update();
reoriented 4D image = tileImageFilter->GetOutput();

This means the total memory budget for the 4D Orient filter is twice the 
size of the input image, i.e. the size of the output image, plus the 
size of all the volumes that comprise the input image. If there was a 
filter to insert an image of dimension D into an image of dimension D+1 
at an offset, then the filter would only use the memory needed to 
contain the output image.

I even know how to write such a filter -- derive from 
InPlaceImageFilter, make the input Image the first input and the volume 
to insert the second input.  I guess my questions are these:  Is there 
enough use for this filter to justify adding it to ITK? And what would 
be a good name for it?