I'm looking for a metadata solution. My environment is Windows.
I'm having trouble finding a definitive answer to your question. This may well be something that has yet to be properly addressed by the relevant standards bodies, and thus something you "can't do" (yet).
That said, here are some pointers to areas for further research. I've investigated them somewhat (though in less detail than is possible, by a long shot), and will give notes on what I've found in each:
I got to this document by way of searches related to things I turned up in other documents, below -- It seems to be saying that the structure for tagging specific areas (or, in the terminology used here, "regions") of a photo is based on a standard from Microsoft that's specifically intended for tagging people. Of course, this doesn't preclude there being some other standard to do what you want... just a data point.
Threads such as this one, showing the (partial) XML content of an XMP file, are how I found the search strings that got me to the above article.
A Python script which also gave some relevant information for strings to search for.
Links to a whole bunch of technical documents (and SDK, etc.) about XMP. In my skimming through a few of these documents, I had trouble finding anything that talked about region-based tagging (even of faces), though it's entirely possible that I was skipping over something because it used words different than what I was searching for.
Another possible Adobe-based source for links to relevant things.
A seemingly-related question on StackOverflow.
EXIF specification document (perhaps EXIF could support it? But I didn't find it there, either.)
The IPTC standard can be downloaded from here, another possible place to look -- though again, I didn't find it there.
Hopefully, this answer is somehow useful to you, and/or will help someone else find an even better answer.
In the mean time, perhaps filing a feature request with the software provider for whatever piece(s) of software you're using, will help drive the changes necessary to make this possible??
Because clearly, it'd be useful. :)
And it has been done, in ways... e.g.: One option is that the Flickr API (and web user interface) offers a way to add "notes", which have specific regions associated with them. If doing the tagging on flickr would be of use to you, then you might try that -- look at the API docs for the various calls that start with "flickr.photos.notes", or (depending on exactly what you're trying to do), just upload a picture, and then click-and-drag across it on its page, and it'll prompt you for adding a note. There may be other programs, sites, etc. that have done this as well, I'm not sure.