SudokuSolverAI

SudokuSolverAI - an app for solving sudoku puzzles through a web camera (augmented reality).

The program detects the sudoku puzzle using the webcam and implements OpenCV for image processing. Using a neural network for digit prediction, it solves the puzzle using an efficient backtracking algorithm. If a solution exists, it showcases the answer on the screen.

Demonstration

Screenshots of solving Sudoku through MacBook Webcam:

Process

Here is the step-by-step how image processing and image detection works in the program using OpenCV and Tensorflow.

• 1 step: Process the image using the gaussian blur, threshold filter to get the grids of sudoku more accurate.
• 2 step: Find corners and crop the image by outer corners.
• 3 step: Detect the digits using the trained model. Detect the grids and identify order of numbers
• 4 step: Use the backtracking algorithm to solve the sudoku.
• 5 step: Show the solution on the empty grids of the sudoku.

Setup

The project is created with:

• Python: 3.11.5
• OpenCV: 4.8.0
• NumPy: 1.24.3
• Tensorflow: 2.13.0
• Keras: 2.13.1

To run this project, first install all libraries mentioned in requirements.txt using the code below:

pip install -r requirements.txt

Then run the project in the main package using the following code:

python3 main.py