Feedbacks about AVM tags in Aladin

[Jonathan Fay]

I have found AVM1.1 to be extremely useful. AVM is for press release quality images, not for simple expression of FITS headers in a exactly corresponding JPEG file. It is assumed some work is going into the images.

Press release images are enhanced, manipulated, mosaic and the published. AVM allows for more than just the coordinates. In contains contact, description, organization, composition of bands, etc. This is much more that what is contained in a FITS header.

While FITS liberator is helps with the import of the image by mapping FITS tags, Photoshop can't track the crops and size changes the user might make. This is not a bug, but maybe an inadequacy of Photoshop for not understanding astronomical coordinate system! Tools like astrometry.Net can help recover correct information for adding AVM tags to processed, scaled and cropped images.

I think XMP writing tools is a weakness in AVM. We have had to develop our own tools to write XMP without Photoshop. For our part we will publish our code for others to re-use, and as others adopt AVM I am sure other code libraries will come to exist and be distributed.

As to CD matrix and embedded FITS headers. AVM 1.1 was created to cure the issues that exist in the ambiguity that is present when different software convert FITS images to JPEG, TIFF, ETC. While the X & Y scale tell a application how to interpret the 2d array in the FITS file, the arrays may be stored upside down or right-side up in a JPEG, TIF, PNG or other format, creating a ambiguity in how to map the image onto the sky with the array format from the image file and FITS header information. I saw an almost equal distribution of methods from several software packages on how this maps.

That ambiguity led to the AVM1.1 creating a non-ambiguous way of mapping the scale, rotation and reference pixels so that the images will always have only one (the correct) way to map to the sky. The result is that every AVM 1.1 images I have seen maps correctly in the Sky in WorldWide Telescopes implementation of AVM 1.1. This is not the case in 1.0, and not the case for FITS headers sidecar images with JPEG/PNG/TIFF files. Additionally a CD matrix allows for non-rectangular and skewed image data. Neither is appropriate for press release images, as there are corrected for in the processing stages of image preparation.

AVM is still new and there are still issue with code library availability, but I believe it is the right direction. I think we should concentrate on better supporting and improving AVM, rather than fragmenting efforts.

Jonathan Fay
Microsoft Research WorldWide Telescope

-----Original Message-----
From: Pierre Fernique [mailto:fernique at simbad.u-strasbg.fr]
Sent: Monday, November 24, 2008 5:25 AM
To: interop at ivoa.net; agauthier at as.arizona.edu
Subject: Feedbacks about AVM tags in Aladin

Dear IVOA members,

We are pleased to announce that Aladin (beta release) is now supporting
the AVM tags in JPEG images (IVOA DRAFT Note 2008 May 14 -
http://www.virtualastronomy.org/AVM_Version1.1May312008.pdf).
Concretely, it means that the Aladin users can click-and-drag a Spitzer
or a Chandra Press Release image from their browser into Aladin and get
a colored astrometrical calibrated images (from the "WCS" information
provided in the AVM tags).

However, we were a little bit disappointed by our experience. We have
several points that we would like to raise to the IVOA community:

1) Presently, the main (unique ?) method for incorporating AVM tags in
the image is Fits-liberator. It is a Adobe Photoshop plugin. This plugin
is free but not Photoshop. This plugin has a surprising behavior (bug
?): it does not adjust the WCS parameters if the user crops, rotates or
resamples the original image. And concretely, the press release images
are often badly registered.
It existed a second method presented in the AVM document as an
Adobe-independent method but unfortunatelly this tool has been
deprecated when Adobe modified its XMP standard (see below)

2) Technically, the AVM tags have to be embedded in a XMP section inside
a JPEG comment segment (or equivalent for TIFF and PNG). XMP (eXtensible
Metadata Platform) is a XML format defined and used by Adobe for
describing image additional information such as EXIF data related to
camera picture metadata. Unfortunately, IVOA has no power for
controlling XMP format. And concretly, XMP is evolving according to the
Photoshop versioning and at the end some press release images used the
old XMP format, and other the new one, or also a mixture of both.
- One syntax : <avm:Spatial.ReferencePixel> <rdf:Seq>
<rdf:li>1044</rdf:li> <rdf:li>983.5</rdf:li> </rdf:Seq>
</avm:Spatial.ReferencePixel>
- other syntax :  <rdf:Description
avm:Spatial.Rotation="0.86206593355939" ...>

3) About WCS. Just the simplest WCS method is supported in the AVM tags.
And for instance the CD matrix which was supported in AVM 1.0 has been
surprisingly deprecated in AVM 1.1. We have presently only a pixel
reference, a sky coordinate reference, a scale factor and a rotation.
The original image size can be also provided in order to compute a bad
ratio factor for bypass an user resampling (but working only if the
reference pixel is the image center - see point 1).
To be honnest, the AVM 1.1 draft is also presenting the possibility for
writing a full FITS header in a dedicated AVM tag but the syntax is not
described (new line ? or 80 columns ?) and concretely, I was not able to
find an example with a such tag. Perhaps because it could be difficult
to put a full FITS header in a simple XML attribut (XMP new syntax).


So, concretly we are a little bit reluctant to extend Aladin for
writting AVM tags in JPEG output images as XMP format can change again
without any IVOA agreement (Adobe has a trademark on XMP, and retains
control over the specification - see the recent example point 1 above).
For my part, I would prefer to save the original full FITS header into a
simple comment segment of the image. It is easy to read (string foo.jpg
on Unix,  notepad2 foo.jpg on windows), easy to generate (jhead or
wrjpgcom tool or any other jpeg tools) and it avoids to re-invent all
FITS keywords (and notably the WCS tags) (also implemented in the cited
Aladin beta release).

Could we compare our experience with some other IVOA members ? Other AVM
reader or writer experiences ?

Best regards
Pierre Fernique

PS. Aladin beta release 5.901 is available via the official Aladin
download page.