more on FITS and VO

[Doug Tody]

Steve -

This is a good point - although a 2D program can and often will default
to the first plane of a 3D image.  If we make this generalization then
we should also permit 1D "images".

FITS WCS makes a distinction between the "image" dimensionality, given by
NAXIS, and the WCS dimensionality, given by WCSAXES.  The WCS can have
a dimensionality greater than or equal to the "image" dimensionality,
which is the number of nondegerate axes (NAXISj > 1).

The question then becomes whether we should permit FITS images with N
nondegenerate axes to have the MIME type image/fits if any of the first
N axes of the image are degenerate.

A related question is whether an image with N nondegenerate axes should
require NAXIS=N (since many programs key off NAXIS to determine the
image dimensionality).  In actual images NAXIS may be anything between
N and WCSAXES.

For example, we could require that image/fits requires that NAXIS=N and
the first N image axes be nondegenerate.  This would deliver "images" that
virtually any program would be able to deal with.  More sophisticated image
visualization or analysis programs which can deal with any image format
could request application/fits, and may or may not get back something
which can be interpreted as an "image".  An alternative would be to allow
image/fits for anything for which the first PHU can be interpreted as an
"image" (as opposed to e.g. a table).

I think this can be specified either way so long as we define precisely
what we mean.  The question is what is the most useful distinction to make
between image/fits and the catch-all application/fits.  If we opt for full
generality than any legal FITS object which can be directly viewed as an
"image" should be allowed for image/fits (for example any legal FITS image
extension extracted to a PHU).  If we argue for a simple basic image model
which won't suprise clients then it makes sense to restrict the model to
a simple N-dimensional array where NAXIS=N.

I'm less sure about this than I was initially.  At this point I think
either approach would work.  At this point we don't have a well defined MIME
type so we are free to define it as we choose.  The main thing is to be
precise and unambiguous.

	- Doug



On Mon, 1 Dec 2003, Steve Allen wrote:

> On Mon 2003-12-01T09:30:24 -0700, Doug Tody hath writ:
> > In short - it is not a simple image unless the first two image axes are
> > nondegenerate.  An actual NULL image is a special case and is not the same
> > as a non-NULL image with primary degenerate axes.  We could live without
> > NULL images (they will cause problems for some programs), but they can
> > be occasionally useful to handle limiting cases using only FITS syntax.
> 
> The current wording of the Internet Draft makes it clear that
> image/fits need not have two dimensions.  This follows the precedent
> set by several pre-existing MIME types.  For example,
> 	image/gif
> admits 3-d images (animated GIFs).  And, appropriate or not, Autocad
> has use of the types
> 	image/vnd.dwg
> 	image/vnd.dxf
> 	image/vnd.svf
> which are actually procedurally-described drawings rather than
> arrays of pixels.
> 
> Significant reworking of the draft text would be required to restrict
> image/fits to two dimensions.  This would probably require discussion
> between the radio and non-radio FITS camps about things like whether
> COMPLEX-valued FITS images are two or three-dimensional.
> 
> Should I subscribe to interop at ivoa.net ?
> 
> --
> Steve Allen          UCO/Lick Observatory       Santa Cruz, CA 95064
> sla at ucolick.org      Voice: +1 831 459 3046     http://www.ucolick.org/~sla
> PGP: 1024/E46978C5   F6 78 D1 10 62 94 8F 2E    49 89 0E FE 26 B4 14 93
> 
>