I wonder how good a job the ClearScan feature on Adobe Acrobat would do. IIRC it creates one or more fonts based the existing characters in the PDF. So each lowercase 'a' would look the same, and be a sort of average of all the 'a' letters in the book.