Vision Based Sudoku Solver

Important stages in the project

• Preprocessing the image to get only the sudoku in the image, so that 81 cells can be extracted easily
• Predicting the digits in each cell using machine learning
• Solving the sudoku using backtracking

Dataset used for training tensorflow based CNN model

Kaggle Dataset which boosted the accuracy to great extent and is much more useful than MNIST in this case

Guide to run

• For solving sudoku: python solve.py

• For training model: python model_train.py train

• For evaluating model: python model_train.py evaluate

Model Summary

Example of a run

• Command: python solve.py --image_path ./assets/sudoku_images/sudoku.jpg --model_path ./assets/models/model.keras
• Sudoku image:

- Terminal result: