> I think this is the kind of thing you can use a filter for

But, it gets more complicated than that. I have XMP metadata (license,
caption, etc.)  stored in the image files as well. And, in order to
export that, I need the path to the source image file. So, my image link
exporter needs the :base-directory to find the source image file. If I
try to get the XMP metadata from the published image file, then I
introduce a race condition with the export of the org file becoming
dependent on the image file already being published.

Perhaps, filters could be useful. I need to think about it. So far, I
have generally stayed away from filters because they can only get the
HTML as text, and have to use some kind of regex to operate on it. HTML
is structured data, and it would have been good if it was available in
some kind of S-form tree structure. It would have saved me the trouble
of parsing the HTML, and I could have used a library like xmlgen to
generate it.

Regards,
Arun Isaac