Handwritten Digit Recognition using Convolutional Neural Networks
This mini-project was undertaken as part of COMP-551 at McGill University.
The goal of this project was to develop machine learning methods for identifying and classifying the biggest digits from a modified MNIST dataset. A simple KNN classifier was implemented as a baseline. Different CNN architectures were investigated in order to get the best possible results.
The following files were used:
data_extraction.py : to load all 64x64 greyscale images
extract_digit.py : to extract the digit that occupies the largest space in the original image
KNN.py : baseline model
CNN_main : final CNN architecture used
See writeup.pdf for details on methodology and results