Just an observation, not a criticism or anything... my biggest concern here is with stitching. As soon as you have the computer applying interpolation between two overlaid images, you're in the position of modifying the original scanned data - both deleting and recreating it, with added filtering in both linearly and possibly rotationally, depending where the software decides the overlay is and how well-designed the software algorithms are.

While the intent is to extract the maximum detail from the scanned image, is fictional data around the joins necessarily the best way to achieve this?

Neil