This repository contains the Matlab code that uses 2 custom built machine learning algorithms (K-nearest Neighbour) to perform Optical Character Recognition. It also includes the EMNIST dataset of 26,000 images and labels on which to train and test the model. A 'summary.pdf' file has been provided to discuss my findings.
The code loads the data in from the EMNIST dataset. It splits the data into 50% for training and 50% for testing to avoid overfitting. This means each model runs 13,000 Images so can take several minutes. The models used are -
- Custom K-nearest Neighbour using Euclidean Distance.
- Custom K-nearest Neighbour using Manhattan Distance.
- MatLab K-nearest Neighbour
- MatLab SVM for Multiclass
All that is needed to run the code is the main.m file and the dataset-letters file.
The code will create a folder called 'Results' in the current directory and create files 'Dataset.png' and 'Confusion.png'.
- 'Dataset.png' - Provides an image of a small sample of EMNIST Images and labels.
- 'Confusion.png' - Provides confusion charts of the 4 models once they have finished running
