As far as
screen readers are concerned, at least the ones with a virtual buffer (JAWS, NVDA, Window-Eyes on Windows), the table doesn't matter: once you're in a form, you're in Forms Mode and all the table elements are ignored, unless you deliberately take yourself OUT of Forms Mode to hear everything (but that means your keystrokes are for navigation and won't fill in the inputs).
If you use alt text for the image, it kinda defeats the purpose of the image (if you were using an image so bots couldn't see the question). So I'd just have it be a regular label instead. If you insist on an image, then yes, use alt text.
The good news is, most of the bots wandering around out there don't have built-in math capabilities to answer those questions, even though it would be super easy for anyone to add. You also get the possibility of making things a bit harder for the cognitively impaired, like people with (I kid you not)
dyscalulia. I like honeypots a bit more for this reason.
You can probably find a lot about honeypots in forms online, but the rule for anyone using a screen reader is generally so: make sure there are instructions in the label that tell humans what the right thing to do is. You can prevent the majority of users from ever running across these honeypots by using Javascript to remove them and send instead the appropriate answer to your server... then only users without JS and bots actually have to deal with the honeypot, and the honeypot question should be easy for the humans.
Bookmarks