Multiple Handwritten Digit Recognition

Recognizing multiple digits from user-drawn input using CNN and GUI

GUI Working Demo

Project Overview

Objective: Build an interactive tool to recognize multiple handwritten digits using a trained CNN model.
Tools Used: Python, Keras, OpenCV, NumPy, PIL, Tkinter

Problem Statement

The aim of this project was to allow users to draw multiple digits on a canvas and have the system accurately extract, preprocess, and classify each digit using a convolutional neural network trained on the MNIST dataset. The solution includes an easy-to-use Tkinter-based GUI.

How It Works

User draws digits on a white canvas using a mouse.
Canvas image is saved and processed with OpenCV to extract digit contours.
Each digit is resized and padded to match MNIST format (28x28 pixels).
Digits are passed to a trained CNN model for classification.
Predictions are displayed on both GUI and original image using OpenCV.

Model Details

Architecture: 2 Conv2D layers + MaxPooling + Dropout + Dense layers
Dataset: MNIST (60,000 training, 10,000 test images)
Accuracy: Achieved over 98% on test set
Loss Function: Categorical Crossentropy
Optimizer: Adam

GUI Functionality

Built using Tkinter, the interface includes a drawing canvas where users can write digits with their mouse.
A “Extract” button processes the written digits and displays predictions.
A “Erase All” button clears the canvas for new input.
The drawn digits are saved as an image, processed using OpenCV (grayscale conversion, thresholding, contour detection), and then passed through the model for prediction.

Results