Sudoku Solver (Computer Vision + CNN)

This project detects a Sudoku puzzle from an image, recognizes digits using a trained CNN, solves the puzzle, and overlays the solution back onto the image.

Project Layout

• sudokuMain.py: End-to-end pipeline that reads an image (optionally from PostgreSQL), detects the board, runs digit recognition, solves the puzzle, and shows the overlay.
• sudokuMain2.py: Alternate pipeline using utils2.py helper functions.
• sudokuSolver.py: Backtracking Sudoku solver.
• utils.py: Image preprocessing, contour detection, digit extraction, and rendering helpers.
• utils2.py: Alternate set of image utilities used by sudokuMain2.py.
• Resources/model_trained.h5: Trained CNN model for digit recognition.

Prerequisites

• Python 3.9+ recommended
• A working camera or image file for testing
• (Optional) PostgreSQL if using the image-from-DB flow in sudokuMain.py

Setup

1. Create and activate a virtual environment.

2. Install dependencies.

pip install numpy opencv-python tensorflow psycopg2 h5py

If you are on Windows and psycopg2 fails, use:

pip install psycopg2-binary

Running The App

Option A: Run With Local Image

1. Put an image named 1.jpg (or update the filename) in Resources/.
2. Update the path in the script:
   • sudokuMain.py uses pathImage = "Resources/1.jpg"
   • sudokuMain2.py uses pathImage = "Resources/1.jpeg"
3. Run either script:

python sudokuMain.py

or

python sudokuMain2.py

You should see OpenCV windows showing the detected digits, solved digits, and the final overlay.

Option B: Run With Image From PostgreSQL

sudoMain.py (typo: sudokuMain.py) includes a helper to fetch an image from PostgreSQL and save it to Resources/1.jpg.

To use it:

1. Ensure your DB has a table with an image byte column (see read_image_from_db in sudokuMain.py).
2. Update the DB connection settings in sudokuMain.py:
   • database
   • user
   • password
   • host (if not local)
3. Run:

python sudokuMain.py

Note: Do not commit real credentials. Use environment variables or a local config file if you need to store secrets.

How It Works (High Level)

1. Preprocess the image (grayscale, blur, adaptive threshold).
2. Find the biggest contour and warp it into a square.
3. Split into 81 cells and run the CNN to recognize digits.
4. Solve the Sudoku using backtracking.
5. Overlay the solved digits back onto the original image.

Troubleshooting

• If no Sudoku is found, try a clearer image with higher contrast.
• If TensorFlow fails to load the model, verify Resources/model_trained.h5 exists and matches your TF version.
• If the prediction threshold seems too strict, check the getPrediction/getPredection logic in the utils files.
• OpenCV windows might not appear in some IDEs; run from a terminal.

Development Notes

• utils.py and utils2.py are slightly different implementations. Stick to one pipeline for consistency.
• The solver (sudokuSolver.py) operates on a 9x9 array of integers with zeros as blanks.

Tests

No automated tests are included yet. If you add tests, consider unit tests for:

• Sudoku solver correctness
• Image preprocessing and contour selection
• Digit classifier output shape and thresholds

License

No license file is included. Add a LICENSE if you plan to share or distribute the project.