OMR Parsing

This is a project to tinker around with OpenCV. It aims to 'parse' OMR sheets and record the information

Currently the OMR sheets are generated using Inkscape. Processing of scanned OMR copies work 90% of the time, the preprocessing of the images still need work.

The OMR sheets contains a QR code which can be used for identification and 2 ArUco markers.

How to run ?

You need to install opencv-python and numpy

Trying with given samples:

Samples are located in the samples directory

# Generating template
python main.py template res/omr.png res/omr.template

# Parsing a sample OMR
python main.py parse res/samples/omr_with_bubbles.jpg res/omr.template

# Parsing a sample OMR with some folds
python main.py parse res/samples/omr_with_folds.jpg res/omr.template

Generating template:

python main.py template omr.png omr.template

Here omr.png is a "soft-copy" of the OMR sheet, omr.template will contain the relative positions of markers on the OMR (QRCode and the 2 ArUco codes) and bubbles.

NOTE: You can print omr.png in a A4 sized paper, fill it out and scan it in 300 dpi (2580x3507) for parsing. Automatic resizing of scanned images is being worked on.

Parsing OMR

python main.py parse image.png omr.template

Here image.png is the scanned OMR image

Example

Given a OMR sheet like this:

running the script gives a output like so:

OMR ID: 1234569420
Exam Code: B2CB1
Roll Code: 6710298
2:  B
3:  A
4:  D
5:  C
6:  A
7:  B
9:  D
10: A
11: D
13: A
14: A
15: D
16: B
17: D
18: C
19: B
20: A
24: B
26: D
28: A
30: C
Marks:
        - Correct:     6
        - Incorrect:   15
        - Total marks: 3

Here OMRID is the parsed value of the QR Code.

The answers are marked using the following marking scheme:

• Correct: +3
• Incorrect: -1

License

Currently licensed under GPL3