Integrating of Image Processing and Number Recognition in Sudoku Puzzle Cards Digitation

Abstract

<p>The aim of Sudoku puzzle is to fill in the blank cells in a square matrix consisting of nine 3x3 blocks with the numbers 1-9 in a way that no number appears more than once in any row, column, or block. We combined image processing, a convolutional neural network (CNN), and a Sudoku game algorithm to automatically place the numbers 1-9 in the blank cells of a Sudoku square matrix. An image of the Sudoku square matrix is first captured using a camera, and then the vertical lines, horizontal lines, and outer frame of the Sudoku square matrix are detected using Hough transform (HT). Based on the OpenCV module, we proposed an image processing algorithm that captures the numbers in the image and calculates the location coordinates of the numbers in the image. We trained the CNN using the MNIST handwritten digit dataset to recognize the numbers in the Sudoku square matrix. Finally, we used the Python programming language to design a Sudoku puzzle backtrace algorithm that automatically deduces and fills in the blank cells in the square matrix. This study provides further understanding of the critical operating principles of CNNs and lays down a foundation for future research.</p> <p> </p>

Rule-based explaining module: Enhancing the interpretability of recurrent relational network in Sudoku solving

“…This means that the most challenging Sudoku puzzles now are those with only 17 known cells. The goal is to fill the empty cells with the numbers 1 through 9, ensuring that each number appears only once in each row, column, and block [28]. An example of a Sudoku puzzle and its solution is shown in Fig.…”