[fitsbits] Spectral FITS -- encoding extraction area

For encoding irregular extraction areas, I recommand NOT using
a list of polygon vertices, but rather an image, which is much
more general and intuitive. So in one FITS file you could
include both your spectra and a "finding chart" image, say of
integer values=0,1,2,3. Pixels where chart values=1 were included
in spectrum 1, etc. A more complicated variation is where pixels are
included in spectra with various unequal weights. Then I would
use a 3-d image with the weights for each spectra in one plane of
the image.

The "pixel mask" facility used in IRAF and some other projects does exactly
this. Bill (Pence) - do you know what happened to that paper we did on
compression in FITS some years back, which included a discussion of pixel
masks? Pixel masks are good for outlining irregular areas and can be
expressed in a compressed from in a binary table.

The purpose of the region stuff in VO is mainly to outline regions on the
sky in world coordinates. This gets compiled into more efficient formats
such as HTM or pixel mask for use by software. This approach is better
for working in "world coordinate" space; the pixel mask approach may be
preferable for describing masks such as extraction masks, exposure masks,
etc. at the level of an instrument or detector. Doug


On Thu, 17 Feb 2005, Walter Jaffe wrote:

For encoding irregular extraction areas, I recommand NOT using
a list of polygon vertices, but rather an image, which is much
more general and intuitive. So in one FITS file you could
include both your spectra and a "finding chart" image, say of
integer values=0,1,2,3. Pixels where chart values=1 were included
in spectrum 1, etc. A more complicated variation is where pixels are
included in spectra with various unequal weights. Then I would
use a 3-d image with the weights for each spectra in one plane of
the image.


_______________________________________________
fitsbits mailing list

http://listmgr.cv.nrao.edu/mailman/listinfo/fitsbits

Thanks to all for these very helpful suggestions and pointers.
We now have much food for thought. I personally like the idea
of using a separate image (or "mask") to describe the spatial
complexities of the spectral 1-D or 2-D data. There may be a way
to combine the virtues of both WCS-space and Detector-space
information that is encoded in an pixel mask image. The Infrared
Science Archive (IRSA), e.g., may be able to create a tool that
reads the image WCS, interprets the masking (detector) code,
and creates an appropriate polygon or HTM tesselation for
the encode sky coverage. In some sense, a pixel mask passes
on the hard work to the next tool down the line. But at least
all of the information is there.

Can someone point me to an example FITS image where something
like this has been done -- namely:

1. fits binary table of 1-d or 2-d spectra
2. 2-d array consisting of spatial "mask" (integer code)
3. optionally, as Jaffe suggests, a spatial plane consisting of a weighting mask

or something comparable. I'd like to see just how it is done
with FITS (e.g., can you mix within one file a binary table and
a 2-D image array? not sure I've seen this before ...)

Thanks again folks,
-tom


Doug Tody wrote:
The "pixel mask" facility used in IRAF and some other projects does exactly
this. Bill (Pence) - do you know what happened to that paper we did on
compression in FITS some years back, which included a discussion of pixel
masks? Pixel masks are good for outlining irregular areas and can be
expressed in a compressed from in a binary table.

The purpose of the region stuff in VO is mainly to outline regions on the
sky in world coordinates. This gets compiled into more efficient formats
such as HTM or pixel mask for use by software. This approach is better
for working in "world coordinate" space; the pixel mask approach may be
preferable for describing masks such as extraction masks, exposure masks,
etc. at the level of an instrument or detector. Doug


On Thu, 17 Feb 2005, Walter Jaffe wrote:

For encoding irregular extraction areas, I recommand NOT using
a list of polygon vertices, but rather an image, which is much
more general and intuitive. So in one FITS file you could
include both your spectra and a "finding chart" image, say of
integer values=0,1,2,3. Pixels where chart values=1 were included
in spectrum 1, etc. A more complicated variation is where pixels are
included in spectra with various unequal weights. Then I would
use a 3-d image with the weights for each spectra in one plane of
the image.


_______________________________________________

Hi Tom -

or something comparable. I'd like to see just how it is done
with FITS (e.g., can you mix within one file a binary table and
a 2-D image array? not sure I've seen this before ...)

You certainly can. This is a multi-extension FITS file. One would use
an image extension or the PDU for an image plane, and a binary table
extension for the encoded pixel mask. Hopefully someone reading this
will have an actual example of such a file they can point to.

Note there is no real standard for compressed "images" like a pixel
mask. The paper I referred to earlier (White, Greenfield, Pence, Tody)
was an attempt to define one. Nonetheless you will find a fair amount
of data around which includes pixel masks of some form or another.

- Doug

Doug Tody wrote:
Hi Tom -

or something comparable. I'd like to see just how it is done
with FITS (e.g., can you mix within one file a binary table and
a 2-D image array? not sure I've seen this before ...)


You certainly can. This is a multi-extension FITS file. One would use
an image extension or the PDU for an image plane, and a binary table
extension for the encoded pixel mask. Hopefully someone reading this
will have an actual example of such a file they can point to.

Note there is no real standard for compressed "images" like a pixel
mask. The paper I referred to earlier (White, Greenfield, Pence, Tody)
was an attempt to define one. Nonetheless you will find a fair amount
of data around which includes pixel masks of some form or another.

- Doug


Ah yes, multi-ext FITS, of course. My mind was focused on mixing spectral
data (say, 1-D table) with a spatial image. I'd like to see example
of something like that.

Thanks Doug for setting me straight. -t

That paper on FITS image compression was presented at ADASS IX, (1999):
http://adass.org/adass/proceedings/adass99/P2-42/
The pixel mask image compression algorithm is supported by CFITSIO, and the
implementation is described at:
http://heasarc.gsfc.nasa.gov/docs/so...mpression.html

It is common practice in astronomical X-ray spectro-analysis to provide a
weighted 2D image map (WMAP) to define the area on the detector or sky that
was used to extract the spectrum. An example of a ROSAT spectral FITS file
is at:

ftp://legacy.gsfc.nasa.gov/rosat/wga...01_in_94.pha.Z

in which the image map (uncompressed) is contained in the primary array, and
the spectrum is stored in the following FITS binary table.

The details of the WMAP format are described in a document at
http://heasarc.gsfc.nasa.gov/docs/he...ip_96_001.html

Bill Pence


Doug Tody wrote:
The "pixel mask" facility used in IRAF and some other projects does exactly
this. Bill (Pence) - do you know what happened to that paper we did on
compression in FITS some years back, which included a discussion of pixel
masks? Pixel masks are good for outlining irregular areas and can be
expressed in a compressed from in a binary table.

The purpose of the region stuff in VO is mainly to outline regions on the
sky in world coordinates. This gets compiled into more efficient formats
such as HTM or pixel mask for use by software. This approach is better
for working in "world coordinate" space; the pixel mask approach may be
preferable for describing masks such as extraction masks, exposure masks,
etc. at the level of an instrument or detector. Doug


On Thu, 17 Feb 2005, Walter Jaffe wrote:

For encoding irregular extraction areas, I recommand NOT using
a list of polygon vertices, but rather an image, which is much
more general and intuitive. So in one FITS file you could
include both your spectra and a "finding chart" image, say of
integer values=0,1,2,3. Pixels where chart values=1 were included
in spectrum 1, etc. A more complicated variation is where pixels are
included in spectra with various unequal weights. Then I would
use a 3-d image with the weights for each spectra in one plane of
the image.


_______________________________________________
fitsbits mailing list

http://listmgr.cv.nrao.edu/mailman/listinfo/fitsbits





--
__________________________________________________ __________________
Dr. William Pence
NASA/GSFC Code 662 HEASARC +1-301-286-4599 (voice)
Greenbelt MD 20771 +1-301-286-1684 (fax)