Handwritten-Digit-Recognition-using-Deep_Learning

This project recognizes the handwritten numerical digits(0-9) that are drawn on the drawing window.

How to use the project

Simply clone the repository and install all the packages in the requirements.txt file and run the interface.py file using python interpreter.

Dataset

The dataset that I used is MNIST as it has a good number of training examples that allowed me to build a powerful and accurate model.

Interface

• The interface is designed using pygame to create an interactive window on which user can draw multiple digits one by one.
• Pressing the key c clears the window.

Details on the model

1. The model is built using Convolutional Neural Networks with a total of 3 layers including the output layer and Dropout(to prevent overfitting).
2. For the hidden layers activation used is 'ReLU' and for the output/classification the activation is 'softmax'.
3. The model is trained using Keras Sequential API with training accuracy of 99.84% and testing accuracy of 98.98%
4. It can clssify most of the digits pretty accurately even with a few irregularities(as shown below).