A question on keywords for 3-D images

[Tom McGlynn]

This sounds like an application for a binary table but
it's not clear whether it should have 32 rows or 32 columns.

If you use the 32 row approach, then each row would have
a field for the 2-D array and additional fields for your
variable 'keyword' information.  You could describe the
geometry of the images using a single set of of keywords.
E.g., you'd have something like
...
    NAXIS2 = 32   / Not trying to make the syntax here perfect!
...
    TTYPE1 = 'DATA '
    TTYPE2 = 'PARAM1'
    TTYPE3 = 'PARAM2'
...
    TFORM1 = '10000F'
    TDIM1  = '(100,100)'  / For a 100x100 image
    TFORM2 = 'F'          / If the parameter is a real...
...
in the header.

If you use the 32 column approach, then the table would only
have a single row with each field being one of your images.
You'd have the advantage of have separate scaling keywords
to scale each image, but at least in principle you'd need
to include geometry keywords for each of the fields.
You'd use field keywords encoding the 'keyword'
information.  This approach may be especially attractive if
the type of the information may be different for some of the
32 planes, e.g., perhaps some planes are binary masks, some
integers and some reals.   

Some fragments of the header might be like:
...
   TFIELDS = 32
...
   TTYPE1  = 'DATA_PLANE1'
   TTYPE2  = 'DATA_PLANE2'
...
   TFORM1  = '10000F'
   TDISP1  = '(100,100)'
   TFORM2  = '10000I'
   TDISP2  = '(100,100)'
...
   PONE1   = 'Value for first param for plane 1'
   PONE2   = 'Value for first param for plane 2'
   PTWO1   = 'Value for second param for plane 1'
...


Either of these formats is reasonably standard, so I wouldn't
anticipate problems with any but the simplest FITS readers.

		Regards,
		Tom McGlynn

bob at ipac.caltech.edu wrote:
> 
> Hello,
> 
> The SIRTF project is designing the format of its output products.  One
> of the products, for example, may contain 32-planes of 2-D data, all
> with the same geometry.  This would normally be a straightforward
> problem, just use a 3-D image, but there is a hitch ...
> 
> There is a set of at least 100 header keywords that have different
> values for each of the 2-D planes.
> 
> There are a number of workable solutions to this and we could just
> pick one and go with it.  However, a standard solution would be
> preferable since it would be better understood and it could be used
> directly in existing software.
> 
> Here are two of the options being considered:
> 
>  1) Append plane number to keywords
> 
>    Use a 3-D image and append the plane number to the keyword names,
>    such as SIGMA01, SIGMA02, SIGMA03.  Is there any *STANDARD* way to
>    do this (especially one that existing software uses) ?  Also, each
>    plane may want to have its own BSCALE and BZERO. Is it possible to
>    specify these values for the individual planes?
> 
>  2) Use FITS image extension
> 
>    Each extension would contain a 2D image and its own set of
>    keywords.  Going to this form loses the 3-D nature of the data with
>    each 2-D image looking like an independent image rather than being
>    part of a 3-D cube.  Also, there is a concern is that many FITS
>    readers and standard software may not be able to deal with files in
>    this form.  It may not be supported well in IRAF, CFITSIO, etc. for
>    3-D operations.
> 
> Thanks for any help,
>  Bob Narron