Handwritten Digit Recognition Using Convolutional Neural Networks (CNN)

This project implements a Convolutional Neural Network (CNN) to recognize handwritten digits from the MNIST dataset, using TensorFlow as the backend.
A simple graphical interface allows users to upload, draw, or capture an image for real-time prediction.

---

🎯 Project Overview

• Built and trained a CNN model to classify handwritten digits (0–9).
• Used TensorFlow and Keras for model creation, training, and evaluation.
• Implemented OpenCV for image processing and checkpoint restoration during prediction.
• Developed a GUI using PyGTK 3.0 for user interaction.
• The system allows users to:
  • Upload an image for prediction
  • Capture an image using a camera
  • Display a stored image
  • Draw digits via gestures for testing

---

🧩 Project Structure

File / Folder	Description
cnn.py	Defines and trains the CNN model, with checkpoint save/restore functions.
train.py	Runs model training and saves weights to checkpoint_files/.
process_image.py	Loads the trained model, preprocesses images via OpenCV, and predicts digits.
main.py	Launches the PyGTK GUI for upload, camera, or gesture-based input.
checkpoint_files/	Stores saved TensorFlow checkpoints.
images/	Contains sample training images and UI background assets.

---

⚙️ Tech Stack

• Python
• TensorFlow / Keras
• OpenCV
• PyGTK 3.0

---

🚀 How It Works

1. The CNN model is trained on the MNIST dataset.
2. When launched, the program loads the saved model weights.
3. Users can provide input via upload, camera capture, or direct drawing.
4. The preprocessed image is passed through the CNN to predict the digit.

---

🧠 Learning Outcomes

• Implemented deep learning with CNNs for computer vision tasks.
• Integrated OpenCV for image preprocessing and real-time predictions.
• Combined backend machine learning with a Python-based GUI for usability.
• Gained experience managing model checkpoints and multi-script ML pipelines.