Due to the COVID-19 pandemic, universities have been moved online only. As a result of that, many of the quizzes which were in person has been moved to online assignments and assessments. To make marking easier, instructors will often force students to divide PDF solution files and upload only the necessary pages to each question. As a result, people who prefer to use pen and paper to write solutions will be at a disadvantage. Although tools for splitting and merging are available, this process can be very tedious and takes a significant amount of time. For professors who impose a low time limit to prevent cheating, spending less time on splitting and merging can be the difference between getting a mark and receiving an automatic 0. With that being said, we present Papertition, the automatic solution for splitting assignments.
What it does
Papertition uses CV and Google Vision API to detect page information. All you have to do to split an assignment for submission is to write an output pdf ID on top of your assignment, add a page number and Papertition will take care of the rest automatically. If there are any pages that were not read properly, manual adjustments can be done via the user interface.
How we built it
The backend of the software uses OpenCV to locate and crop the PDF ID and page information. The cropped image is then fed into Google Vision for handwriting recognition. The GUI is built using Swing in Java.
Challenges we ran into
One of the main issues was that Google Vision often gives too much information (especially on math assignments). To reduce the amount of information Google Vision gives, OpenCV is used to preprocess the images and only send specific areas of interest.
Accomplishments and Lessons
We were able to create a complete program with a GUI. The program is ready to be used and tested by the general public without requiring any knowledge of programming. It is able to recognize and automatically process most handwriting without any issues.
What's next for Papertition
For the future, we plan to use less invasive markings to ID pages. With our current setup, although the markings are out of the way, they stand out. Ideally, the marking should be invisible to prevent confusing the marker.