I would just put the caption before or after the image, and wrap both in a block with "page-break-inside: avoid" to disencourage the reading system from inserting a page break between image and caption (but not really prevent it).
If the caption is very long, I'd even insert pagebreaks before and after both caption and image, like in Kipling's
Just So Stories. Basically you find a page with only the caption ("The following illustration shows this and that..."), then a page with only the illustration to its full glory, then the story continues.