akshitsarin / handwriting-recognition-using-convolutional-neural-networks

Prediction of User-Drawn Digits using Convolutional Neural Networks through Python. Created using flask. Allows the user to draw a digit and predicts the number using the Keras API and Tensorflow.

Handwriting-Recognition-using-Convolutional-Neural-Network

Prediction of User-Drawn Digits using Convolutional Neural Networks inside a web-app, created using Flask MicroFramework for Python.
The web-app allows the user to draw a digit on a virtual canvas and predicts the drawn digit using Keras-API and Tensorflow in python.

Updated README:

As of May 2020, more than half of the functions are either out-dated / depreciated. The project was created in 2017, and the supporting library versions for the dependencies are as follows:

tensorflow: v0.10.0
keras: v1.1.0
scikit-learn: v0.19.0
numpy: v1.14.1

Structure:

The project uses the MNIST database of handwritten digits, with normalized sizes. The dataset is divided into 4 parts - Training Images, Training Labels, Test Images and Test Labels.
The keras model created is stored in the model.h5 file (HDF5 type). For each prediction, the model is loaded into the flask web-app through the load.py file inside the model folder.
The pattern drawn on the canvas is converted into output.png. app.py loads the image, applies normalization (converting the image into 28x28 pixels), which is then passed into the predict method of the model. The value is then returned to the index.html page.