MNIST - Handwritten Digit Classifier with GUI

This is a deep learning project where we have classfied hand written digits using simple ANN.

GUI is made using python Tkinter where we can draw a digit, upload a photo and also paste the link of the image consisting of a digit and ANN model will display the predicted digit.

Dataset - tensorflow.keras.datasets.mnist

MNIST is a dataset of handwritten digit images consisting of a training set of 50,000 images and a test set of 10,000 images. Each example is a 28x28 grayscale image, associated with a label from 10 classes ( 0 - 9 )

Experiment Results:

• Performance Evaluation
  • Out of 50000 training images 10000 images were used for validation set.
• Training and Validation
  • Dense ANN with 2 layers is used with RELU as activation function and softmax for output.
• Performance Results
  • Training Score: 100.0%
  • Validation Score: 98.03%
  • Testing Score: 98.0%

GUI Features :

• Draw a digit - Tkinter canvas is used to draw
• Upload a photo - Upload a photo of handwritten digit
• Paste a link - Paste a link of the photo includes handwritten digit
• Clear - Clear previous drawing and sketch new
• Save - Save the image

Demo

First Look

Draw and Predict

Save Image

Saved Image

Predicting saved image

paste link of Image and predict

Further Improvements

There are lot of things to improve upon

• CNN can be used to increase accuracy on unknown data.
• More data can be provided.