Implementation of handwriting recognition using machine learning.

• Python 3.6
• TensorFlow 1.6
• SciPy 1.0
• Keras 2.1
• EMNIST dataset

Before training, ensure that emnist-byclass.mat from the EMNIST dataset is stored within the data/ directory.

Run the training script using:

python3 train.py

A model h5, YAML, and Pickle file will be generated within the model/ directory.

Once the model has been created, a handwritten character images can be predicted by running:

python3 predict.py [file1] ...

Note: multiple files can be predicted in one instance and a wildcard * can be used to process all files within a directory.

A Tkinter GUI with a drawing canvas has been created to test the model implementation more easily. This can be started with the command:

python3 draw_gui.py

• GUI testing environment for live demo
• Train on byclass dataset
• Improve prediction accuracy (should improve after switching to byclass)
• Compare byclass with other EMNIST datasets
• Improve image preprocessing
• Train using other learning algorithms (if time permits)
  • SVM
  • Random forest
• Image process strings of letters (if time permits)