sudoku-solve-ai

A summer project made using tensorflow's keras API and opencv library. The program detects a sudoku from an image and solves it accordingly.

Getting Started

Run the following commands in your windows terminal to get the project running locally. Make sure your python version is between 3.6-3.8.

git clone https://github.com/dwij2212/sudoku-solve-ai

Create a virtualenv and activate

python3 -m venv myvenv
myvenv\Scripts\activate

Run this to setup the dependencies in your virtual environment

pip install -r requirements.txt

Usage

Go to the program main.py and change the following line in order to use your own image for recognition.

#main.py

...
    img = cv2.imread(PATH_TO_YOUR_IMAGE, 0)
...

Pass the path to your image in the argument of imread. Now run the following command in your terminal to run the program.

python3 main.py

Modify how your program works

In the file main.py, make the following changes to capture image from your camera.

#get image from camera or local path
#img = cv2.imread('sudokuimg.jpg', 0)

#uncomment this line and comment line above this to capture image from your camera and solve
img = capture()
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

Upgrade model

What upgrading your model basically means is fitting your model on the data fed to program during runtime. Make sure all data fed to the model is correct i.e. the sudoku is printed correctly before training or else model will become inaccurate. This can be enabled by making the following change in main.py

UPGRADE_MODEL = True

Since this step is computationally expensive for large dataset, it is False by default.

Limitations

Limitations are listed as below. If you have any suggestions then feel free to submit a PR.

1. The program can identify sudoku grids only if the image contains well defined borders of the outermost square of the puzzle.
2. The program assumes that the puzzle occupies ample space in the image.
3. Program can detect only 1 grid if more than 1 puzzles are shown to it.