January 24, 2020

274 words 2 mins read



Image processing and computer vision model to automatically evaluate and grade handwritten mathematical equations

repo name divyaprabha123/Autograding-handwritten-mathematical-worksheets
repo link https://github.com/divyaprabha123/Autograding-handwritten-mathematical-worksheets
homepage https://towardsdatascience.com/computer-vision-auto-grading-handwritten-mathematical-answersheets-8974744f72dd
language Jupyter Notebook
size (curr.) 9414 kB
stars (curr.) 40
created 2019-12-05

Auto-grading of handwritten mathematical worksheets

This repo is the part of internship project with Bosch/RBEI-EDS2. Aim of this project is to digitize the steps of solving a mathematical equation written by freehand on a paper, validating the steps and final answer of the recognized handwritten lines by maintaining the context.


alt text As shown, the overall solution can be divided into the following two parts:

  • Workspace Detection module
  • Analysis Module

Workspace Detection module is responsible for detecting multiple workspaces in a given sheet of paper using pre-defined markers.

Analysis module is responsible for detecting and localizing characters in lines in any given single workspace, and mathematically analyzing them and then drawing red, green lines depending upon their correctness.

For more detailed description on the workflow see Report.pdf


Each line is corrected separately

  1. Green Box represents - Line is correct
  2. Red Box represents - Line is incorrect alt-text

Target algebraic equation A * X2 + B * Y

Where A = 56, B = 7, X = 3, Y = 13

Line No Equation written Expected Ans Actual Ans Status
1 56 * 32 + 7 * 13 595 595 Correct
2 56 * 7 + 92 595 484 Incorrect
3 595 + 92 595 687 Incorrect
4 595 595 595 Correct

For more detailed description on the workflow see Report.pdf

To evaluate and test


Character Recognition

Use this link to download the DCCNN model for OCR part of the analysis pipeline

comments powered by Disqus