Seems trivial to diff multiple screenshots to identify what parts move. Or just use a compression algorithm to do the same.