[fitswcs] Detector distortion correction representations in FITS

[Richard Hook]

Hi Mark,
Thanks again for all the information. I now understand the problems better
and can appreciate why some options, which superficially seem easy, are in
fact very tricky for the standards!

Mark.Calabretta at atnf.CSIRO.AU said:
> OK, having thought about it some more I think I understand your 
> problem a bit better.  You're reading out lines from a CCD which is 
> spatially distorted in some way and you want to de-distort the pixel 
> coordinates before anything else.  (It would have helped a lot if 
> you'd briefly described your instrument - my background is in 
> radioastronomy.) 
Sorry, yes I should have explained better. What you say is quite right - the
distortion is tied to the detector grid of pixels and is measured in pixels
one way or another rather than on the sky. So, if you rotate the camera the
distortion rotates too, on the sky, but not on the pixel grid.

I think your initial suggestion is probably the best way for us to go in
the short-term: combine the current distortion polynomial information with
the current CD matrix into PV coefficients and set the CD matrix to the 
identity. In other words convert to your proposed TAN-poly projection.
I think software could, should it be necessary, extract the linear
component of such a distortion from the PV coefficients if desired.

I think your discussion of a flexible scheme which has more components (eg,
a polynomial in "pixel space" first, then the CD, then a projection which
may also have additional polynomial coefficients) shows that that approach
really is much too complex and we need to keep things manageable. I think I
would also reject the pre-poly/CD/post-poly suggestion for the same reason.

So, yes, I think I am convinced that your initial suggestion (despite my
list of its deficiencies) is the way to go in the interest of a cleaner
standard.

It is very interesting how the area in which one works colours ones view on
this - to me the polymonial pixel correction/CD/projection seems very natural!
In an ideal world I would, I suppose, like TAN-pixpoly where the PV 
coefficients
are applied before the CD matrix. The "pixpoly" coefficients would then be
decoupled from the projection actually. However, I can see from your comments
that this doesn't really fit in and I am not going to propose it.

Just for information I will forward Doug Mink's message to which I referred
to you (Mark) separately.

Thanks again for taking the time to discuss these points.

It would be good to get comments from others as I am now convinced that this
is a valid problem to discuss even if there is no quick and clean solution.

Cheers, Richard