Overview / Usage

An AI system to identify the sudoku and then solving it using an algorithm. Making the machine to solve a puzzle like sudoku could be a great step towards building smart systems. with the increasing advent of augmented reality, it could help us in solving complex challenges of computer vision and can be used as an automation tool.

Methodology / Approach

From reading the image to showing the solved sudoku back on the image, there are various tasks involved in working in a pipeline.

1. Load Input Image
2. Locate wherein the input image the puzzle is and extract the board using image filtering, corner detection and perspective transform.
3. Given the board, locate each of the individual cells of the Sudoku board (most standard Sudoku puzzles are a 9×9 grid, so we’ll need to localize each of these cells).
4. Determine if a digit exists in the cell, and if so, recognize it using the CNN Model.
5. Apply a Sudoku puzzle solver/checker algorithm to validate the puzzle and solve it.
6. After getting the solved puzzle, display the output result to the user using inverse perspective transform.

Technologies Used

OpenCV for image processing-related tasks for puzzle detection, segmentation, projection on the image, etc.

PyTorch for model training of CNNs for digit recognition.

Python and Jupyter Notebooks. Github for Version Control.

Hardware: Intel Core i5 8th Gen processor, Nvidia MX230 2GB graphic card.

Repository

https://github.com/lavish619/AR-Sudoku-Solver-Pytorch-OpenCV