If graphics aren't labeled, then, of course, you may have a hard time determining what graphic to click.  But the question, as posed, assumes that.  Can NVDA be made to announce all graphics?  I don't think it does now.  If there is a graphic without any text, I don't think it announces it and there is no setting I know of to adjust the level of graphic announcement.  JAWS has such a setting and you were discussing that in your comparison of NVDA and JAWS.  If I am correct that there is no setting in NVDA for this, this is another way in which NVDA isn't as user definable as JAWS where it should be.  This may be a serious problem on a site where, to take an action a graphic needs to be clicked.  It might mean having to click it with the simulated mouse but to do that, it must be announced as you move through the page so you can move to it and then move the mouse to it. 

You may need sighted help to know which graphic to click but as far as I know, NVDA doesn't read unlabeled graphics on web pages and moving character by character doesn't read them either.  As far as I know, they are skipped as though they don't exist.

Gene