Smart Sudoku

A Smart Sudoku that provides the solution to a classical 9X9 Sudoku from its image using OpenCV and Deep Learning.

• Image Recoginition is done using OpenCV using techniques like Blurring, Adaptive Threhsolding and Inversion.

• A digit recoginition CNN model was trained on MNIST dataset.

• The model recoginises the digit from the image.

• Another CNN model was used to find the solution.

• The second CNN model was trained on 1 million Sudoku Dataset which can be downloaded from Kaggle

Stages

1. Stage 1 Cropped Image

1. Stage 2 Image after Blurring, Thresholding, Inversion and Grid Detection

3. Stage 3 Image after Digit Reciginition

4. Stage 4 Final Solution

Steps to run

1. Clone the project

   git clone https://github.com/ishpreet-singh/smart-sudoku
   

2. Move to project directory

   cd smart-sudoku
   

3. Create a Virtual Environment named venv. Read more about Virtual Environment

   virtualenv -p /path/to/python3 venv
   

4. Activate the virtual environment

   source venv/bin/activate
   

5. Run the Project

   python sudoku.py <IMAGE_PATH>
   

   Note: You need to provide the path of your sudoku image. Eg: python sudoku.py dataset/sudoku2.jpg

Project Structure

- smart-sudoku/
   - LICENSE
   - README.md
   - assets/
     - stage1.png
     - stage2.png
     - stage3.png
     - stage4.png
   - controller/
     - __init__.py
     - cnn.model
     - cnn.py
     - digit_recoganise_controller.py
     - gpus.py
     - image_controller.py
     - model.h5
     - model.json
     - sudoku_controller.py
     - sudoku_solver.py
   - dataset/
     - mnist.npz
     - sudoku.csv
     - sudoku1.jpg
     - sudoku2.jpg
     - sudoku3.jpg
   - requirements.txt
   - sudoku.py

File Description

• cnn.model: cnn.py saved weights

• cnn.py: Defining CNN Model for finding Sudoku Solution, given unsolved sudoku

• digit_recoganise_controller.py: Defining CNN modle for recoginizing digits from grid images

• gpus.py: Helper file required by Tensorflow

• image_controller.py: Defining Class for Image reconition using methods like Blurring, Thresholding, Inversion

• model.h5: digit_recoganise_controller saved weight

• model.json: digit_recoganise_controller saved weight in JSON format

• sudoku_controller.py: A backtracking approach to solve Sudoku

• sudoku_solver: Class to solve sudoku using cnn.py.

• sudoku: Main File