What architecture to choose for a neural network that solves the problem of comparing two test forms in png & jpeg format?

Question

Neural networks

Hello!
Task: A system for intelligent determination of the results of test tasks and their evaluation.
Input data: 1 photo is all True results, 5+ photos are random.
Each test task consists of 4 answer options.
At first glance, the task seemed not difficult, just compare the matrix (correct answers) with the
matrices (student answers), I looked at several options, but I can’t choose the most optimal one for this task.
I ask you to advise algorithms, networks, etc., which, in your opinion, can easily cope with this task.
I was looking at options for using binary neural networks or Bayesian networks.
Thanks in advance for your replies.

Answer 1

First you need to recognize.
1. Use the fragmentation of the form into an invisible grid of squares.
2. Position the content of the form so that when choosing answers, the coordinates of the squares with the number of the question and the answer options for it are known exactly.
3. Make some guide marks (bold corners, horizontal/vertical lines, etc.) on the corners of the form sheet (for easier and more accurate recognition)
4. Connect any library for recognizing the finished photo/scan of the form. For example OpenCV.