[Insight-developers] Creating large images

[Richard Beare]

Hi,
Sorry if I'm being very thick, but I haven't been able to find out how
to do this at the filter level - maybe it is only possible if I write
a new filter.

Here's the problem. Basically I'd like to assemble a large image that
is going to be the start of a pipeline. I'll know how big the image
will be, so I can create it in advance. My problem seems to be getting
a pre-allocated image to act as a mapper.

Here are some schematics of the sort of structure I have in mind:

  -----         ------    ...     -------      ------       : n small images
    |               |        ...        |              |
    |               |                   |              |          :
pre processing pipelines
    \               |                   |             /
       --------------------------------                   : One large image
                             |
                             |                                    :
Further processing

Where the pre-processing pipelines may be logically parallel or
created and run serially.

I had hoped to do this be allocating the large image, setting the
requested region to match the size of the small image and have the
start index I wanted and then connect it to the end of the
preprocessing pipeline. However I can't see any way to make a
pre-allocated image act as a mapper.

One alternative I can see at the moment is to pad each small image
with zeros so that they have the desired output size and the input
image is in the correct and then accumulate them, which seems a fairly
round about and inefficient way to go.

The registration framework will probably need to deal with similar
problem if it gets used to assemble mosaics.

So, what am I missing here? Is a new filter necessary, or is there
some other way to put the output of a pipeline in a specific location
in a preallocated image?

All advice welcome.

Thanks