Handwritten Mathematical Expression Recognition System. The Project is built using Keras library.
The dataset used in the project contains 2258 instances of labelled handwritten mathematical equations in InkML format.
The model has been trained on Image data which was obtained by conversion of this InkML data to 28,173 images of 75 classes of unique mathematical expression.
The model achieves an accuracy of 94.29% on the validation data.
The accuracy achieved is pretty decent considering the uneven distribution of images in various classes:
inkml2img.py: The code for extracting traces of different expressions from a given InkML format equation.
convert.py: Converts InkML files to images.
run_model.py: Contains code to build the Deep Learning architecture and train the converted images on the Deep Learning model.
training_monitor.py: It monitors the training based on the validation loss to avoid overfitting.
InkML to Image conversion credits: https://github.com/RobinXL/inkml2img