sudoku-scanner

A scanner that detects, recognizes and solves sudoku puzzles from real-life pictures using a pipeline of classical Computer Vision algorithms.

How to use

Install dependencies (use Python 3.6):

pip install -r requirements.txt

Run on image:

python src/sudoku-scanner.py <path to your image>

For some more options, use:

python src/sudoku-scanner.py --help

Pipeline

The pipeline is divided into 5 main steps

You can generate the following images for your own image by following the steps in /notebooks/VisualizePipeline.ipynb

Input

1. Preprocessing input image

2. Exposing the grid

3. Finding corners and transforming perspective

4. Preprocessing grid

5. Digit cleaning and classification

The individual digits are then cut from the grid, cleaned individually and classified by a small ConvNet.