Overview

This is a simple project that uses a one layer neural network to estimate the boolean function that removes Salt and Pepper (S&P) noise from QR code images, which are binary. It was done for the Machine Learning for Computer Vision course @ the São Paulo University (MAC-6914).

Running the project

1. The QR codes are generated randomly in create_dataset.py.
2. The training image pairs are then transformed into a tabular dataset in preprocess_dataset.py.
3. The neural network architecture is defined in neural_network.py.
4. Training is done in train_model.py
5. Finally, results and image predictions are done in generate_results.py

Requirements

• numpy
• skimage
• pytorch