Anonymization tool for testing
Scan quality varies greatly, and while unit testing will allow us to improve the algorithms, having a source of real data would be even better.
Unfortunately real data is usually hard to obtain due to privacy concerns. We can work around this problem if we implement a helper function (not necessarily exposed via web interface), where we blank out parts of the images that we aren't using in image processing. This way we would be able to rapidly assemble and renew real life datasets. If we additionally read off from the database which pages failed, we will be able to right away focus on the problematic issues.
Later we could use similar tools to for example blank out the student name in the version of pages shown to the grader.