On Sudoku Problem Using Deep Learning and Image Processing Technique

Abstract

Sudoku has been a challenging puzzle for many people, in fact it is considered as a brainstormer. Many find it difficult to solve the puzzle. We generally apply different math techniques to solve the Puzzle. It would be great, if a computer could solve Sudoku. We have used deep-learning concepts to solve the Sudoku puzzle in a most optimized manner. Moreover, image processing and object localization algorithms are used to detect the Sudoku puzzle. Initially, an image consisting of Sudoku is provided, by using object localization & CNN algorithms we detect the Sudoku square box. A 9 × 9 square box is detected. By using ANN, we detect the numbers and empty spaces, later Backtracking algorithm is used in solving the puzzle in a more efficient manner. Puzzle is detected and digits are recognized, which can be digital or handwritten and finally the puzzle is solved. We achieved an accuracy rate up to 99% in detecting the puzzle.