This is a small project which performs optical character recognition using principle compoenent analysis to extract feature fectures, then uses k-nearest neighbours to classify a character in a given bounding box.
The code written by me is in system.py. All other code is used for setup and evaluation purposes.
- Ensure that you have Python 3 installed.
- Clone this repository locally.
- Navigate to the root project directory and install the required packages (in requirements.txt) using pip (
pip install -r requirements.txt) - (optional) Run the training in order to update the 'model.json.gz' file (in '/data') using
python train.py. - Run evaluation on the model using
python evaluate.py dev(this takes a little while, be patient!).
The results will show the acccuracy of the classifier (percentage of correctly classified characters) on each page within the '/data/dev/' directory. You can view the pages in pdf format within this directory.