Sudoku Solver using OpenCV, Simulated Annealing and Python

Hello friends,

In this project, Sudoku is extracted from image (newspaper) and then solved using Simulated Annealing (a method of Numerical Optimization).
The complete code is divided into four parts. 1. Preprocessing 2. Sudoku Box Extraction 3. Cell Digit Recognition 4. Sudoku Solving.
Steps of each part are given below

Pre-processing:
1. Load Original image.
2. Convert to grayscale image.
3. Apply Gaussian Filter to remove noise.
4. Convert Grayscale image to Binary Image.
5. Perform Morphological Closing.
6. Remove image border connected blobs/objects.
7. Perform Morphological Eroding.
8. Get largest blob.

Sudoku Box Extraction
1. Get all corner points.
2. Select only four from detected corner points which are closest to corner points of original image.
3. Sort extreme four points in a particular order.
4. Read Template of Sudoku grid image.
5. Assign corners of template image in same order.
6. Perform Perspective Transformation.
7. Crop Sudoku box image from stitched image.

Perform OCR
1. Get approx. cell size.
2. Crop each cell from Sudoku box image.
3. Preform few additional steps such as noise removal, border connected blob removal.
4. Get Cell with provided digits only and put 0 in empty cell.
5. Use paddleocr library for OCR tasks and predicted digit in cell.

Solve Sudoku
1. Create an initial sudoku.
2. Use Simulated annealing method to solve sudoku.
3. Draw solution to template image.

I hope you enjoyed this video. Feel free to contact me at [email protected]