Indeed, I also can use my usual converter to save in html5 (with images in a separate folder) and then use Calibre to create an EPUB. I did try it. But then, as the EPUB is not exactly built up to my taste, I still have a lot of customisation to add (not only fonts).
I manually found the solution for this book (after some splitting). One image (out of 35) had a faulty src, a leftover from the "original" scanned html. But a preview of the src tags would have saved me the day and I would appreciate to have it handy for the next time.
I find strange that there seems to be not obvious solution for this kind of search. A parser of some kind?
