Re: TASS image compression

[Stupendous Man <richmond@stupendous.cis.rit.edu>]

  Robert wrote:

> I have been looking at the problem of image compression ...
> So I've decided to document and experiment a little bit
> with the data and programs which may be available.

  I always enjoy reading about experiments.  Choosing
hcompress is good, since it's still alive and used widely.

> With scale factors of 1, 2, 4, 8, 16, 32, 64 & 128,
> the hcompress approach (actually fcompress script) reduces
> the normal TASS FITS image from 8,415,350 bytes (~8 MB)
> to a fractional size as follows:
>  1: 0.48, 2: 0.46, 4: 0.41, 8: 0.35, 16: 0.29, 32: 0.23,
>   64: 0.16, 128: 0.10

  Okay.  Those are good numbers to know.  It appears that
the gain in shrinkage grows slowly; most of the shrinkage
appears in the least lossy compression.  That's nice.

> The question then becomes where, if one uses a potentially
> lossy compression strategy, a reduction in the image information
> begins to significantly impact the accuracy of the data.

  Indeed.

  May I ask you, Robert, to continue your work?  There are
several available programs which measure the instrumental
magnitude (or brightness, take your pick) of stars in a
FITS image.  Could you do something like the following:

        - take a sample FITS image
        - pick 10 or so stars, some bright, some faint
        - measure their instrumental magnitudes
        - compress, then uncompress the image with
                 one of the lossy schemes
        - re-measure instrumental magnitude
        - compare pristine vs. compress/uncompress values
         
  If you can find a breakpoint at which the lossy compression
causes errors of, say, 3 percent, or 5 percent, in the output
magnitudes, we'll have a decent rule of thumb for anyone in
the future who wants or needs to compress the data.

                                     Michael Richmond