For annotations, wouldn't make sense to build on the
EPUB Canonical Fragment Identifier (epubcfi) Specification?
Using that to save the location of an annotation makes it easy to always have the correct location even if the book is loaded onto a different reader that flows text differently (screen reader size, font change, whatever).
Combine the above with an XML wrapper to hold the text of the annotation, or the location of the image, and it should be doable, no? Having people agree to what the XML wrapper should be would be another story...