Finding solutions for Sudoku's online can be cumbersome due to tedious text inputs. We want to make this process easier by recognizing untouched and partially completed Sudoku's from photographs and solving them.

What it does

Recognizes Sudoku's from photographs and solves them. Implementing web-UI to allow for custom user input.

How we built it

Image recognition and classification in MATLAB. Solution algorithm and logistics in Python. Uses parts of Chars74K dataset:

Challenges we ran into

Classification of numbers in a noisy and non-ideal environment.

Accomplishments that we're proud of

Functioning solver algorithm. Good performance. Decent recognition. Acceptable classification rate.

What we learned

MATLAB can be faster and more convenient than OpenCV.

What's next for Swagdoku

Improve classifiers with larger and/or more fitting training data. Improve UI. Make solving more reliable and better in giving feedback on solvability/mistakes in the Sudoku.

Share this project: