Sudoku Solutions from Image: Computer Vision

A program that detects sudoku puzzle from an image and solves it using recursion.

The project is divided into three parts:

• Part One: Digit Classification Model
• Part Two: Detecting And Reading The Sudoku From An Image
• Part Three: Solving The Puzzle

Part One: Digit Classification Model

In this section:

• Loading Data
• Splitting The Test Train And Validation Sets
• Preprocessing The Data
• Model Building And Training

Part Two: Detecting and Reading the Sudoku from an Image

In this section:

• READING THE SUDOKU PUZZLE
  • Read an image from the dataset
  • Preprocess the image
• DETECTING CONTOUR
  • Detect the biggest contour of the image.
  • Reshaping the outline to get the cropped and well-aligned Sudoku
• SPLITTING THE CELLS AND CLASSIFYING DIGITS
  • Splitting the sudoku box into 81 cells with empty spaces or digits
  • Cropping the cells to avoid misdetection of boundary lines as digits
  • Using the model to classify the digits in the cells such that the empty cells are classified as zero
  • Getting the detected output in the form of an array of 81 digits

Part Three: Solving the Puzzle

In this section:

• Reshaping the array into a 9 x 9 matrix
• Solving the matrix using recursion

Built With

• TensorFlow 2
• Keras
• OpenCV