Scripts to read a sudoku from an image, detect the filled in numbers and complete the sudoku.

• OpenCV for reading and processing image data
• NumPy for mathematical, in particular array operations
• Matplotlib for displaying differents steps of the sudoku image processing
• Tensorflow for training and re-loading the CNN model
• Scikit-learn for evaluating the training of the CNN model
• warnings for warning if sudoku shape might be too large and lead to inconsistent placements of the digits

Class for reading a sudoku from a file. The largest quadrilateral contour is chosen as sudoku. In this contour, connected components are searched for. If the connected component satisfies certain shape criteria, it is assumed to be a digit. The digit is then classified using a neural network.

Class for solving a sudoku. A backtrack algorithm is used.

A script for training a CNN for number classification. We use the pre-trained CNN model-OCR.h5. It can be run using the tensorflow/keras library:

import tensorflow as tf
my_model = tf.keras.models.load_model('model-OCR.h5')

To use the script, use the main.py script. To do so, set the parameter of the run configuration of the main.py script to use a $FilePrompt$:

When you run main.py, you are asked to choose an image. If you want to run the script from the command line, one has to hand over the path to the image:

python3 main.py /path/to/image.jpg