If I were you, I would do OCR on the text, so that you transform it into a text file, and use Calibre to format it to ePub. There are many OCR software that are free or/and open-source that could do a good job as long as your scanned pages are of good quality. See
http://en.wikipedia.org/wiki/OCR_software for a rather comprehensive list.
Note: Tesseract seems to give excellent results on bitonal image scans, as seen on this article:
http://www.mscs.dal.ca/~selinger/ocr-test/
Hope this helps.