Sudoku Scanner And Solver

Introduction

This is a Sudoku Puzzle Solver application which is built in Qt Framework, PyGame using Python language. It provides an option to either identify the puzzle input by scanning a well cropped puzzle image or by entering the values manually. After automatic recognition of values, the program asks the user to verify the detected numbers and then continue with the solution.

Contents

• Demo
• Installation
• Usage
• Working
• Scope For Improvement

Demo

Installation

• Install 3.8 from here
• Install git from here
• Run git clone https://github.com/DiAnant/Sudoku_Scanner_And_Solver.git
• Run pip install -r requirements.txt

Usage

• Execute main.py
• Upload an image or input the values manually.
• Press enter to solve the board.
• Press 'space' to reset the board.
• Click the particular square in the sudoku grid and press a number key to change its value to that number.

Working

• The app interface is built on PyQt5 (Qt) Framework and the grid puzzle is built using PyGame.
• Tensorflow is used for detection of numbers from the input image. Weights trained on a Deep CNN on MNIST database with very high test accuracies are already prepared.
• A standard backtracking procedure is used to solve the puzzle once the detected numbers are confirmed by the user.

Scope For Improvement

• The current application requires the input image to be perfectly well cropped. An incorrectly cropped or straightened image will result in mis-detection of the numbers.
• A much faster Numpy based Inference algorithm can be developed instead of using tensorflow for detection purpose.
• Instead of using a standard backtracking algorithm, a faster sudoku solving algorithm developed by Peter Norvig can be used.