Welcome to the Digit Recognition System, a Django-based web application crafted to recognize handwritten digits with precision and efficiency. This innovative application bridges the gap between machine learning and user interactivity, allowing seamless digit recognition through a custom neural network model integrated into an interactive web interface.

---

📜 Table of Contents

1. Overview
2. Features
3. Screenshots
4. Getting Started
5. Technical Details
6. Usage
7. Dependencies
8. Contributors

---

📜 Overview

The Digit Recognition System is designed to recognize and predict handwritten digits in real-time. Users can draw digits on an interactive canvas, which the system processes using a custom-trained neural network built from scratch. This project demonstrates the integration of model training, prediction, and deployment, providing a seamless user experience powered by the MNIST dataset and Django framework. It also showcases the potential of machine learning in interactive applications.

---

✨ Features

Custom Neural Network 🧠

• Implemented and trained from scratch using backpropagation with gradient descent.
• Built on the MNIST dataset, which consists of grayscale 28x28 images of handwritten digits.

Digit Drawing Canvas 🖌️

• Interactive web interface allowing users to draw digits for immediate prediction.

Real-Time Predictions 🚀

• The system predicts drawn digits as soon as the user finishes drawing, providing instant feedback.
• Uses the trained neural network for inference directly in the web application.

Neural Network Visualization 🔍

• Visualize the neurons and their connections in the neural network.

---

📸 Screenshots

---

🚀 Getting Started

1. Clone the Repository:

git clone https://github.com/USMAN-FAIZYAB-KHAN/Digit-Recognition.git
cd Digit-Recognition

2. Install Dependencies:

Ensure Python and pip are installed. Then, install the required dependencies using pip:

pip install django numpy tensorflow

3. Train the Model (Optional):

If you'd like to retrain the model with updated settings or hyperparameters:

python neural_network.py

The model will be trained using the MNIST dataset, and a saved pickle file (digit_recognition_model.pkl) will be generated for recognition. This pickle file stores the model's weights and biases, allowing for easy loading and use in predictions.

4. Run the Server:

Start the Django development server:

python manage.py runserver

5. Access the Application:

Once the server is running, open your web browser and navigate to:

http://127.0.0.1:8000

---

🗝 Technical Details

The Digit Recognition System employs a custom neural network with the following components:

1. Network Architecture

• Input Layer (784 neurons): Processes 28x28 pixel images flattened into 784 values.
• Hidden Layer (250 neurons): Extracts features using the sigmoid activation function.
• Output Layer (10 neurons): Predicts digits 0–9.

2. Core Algorithms

• Sigmoid Function: Enables non-linearity and gradient calculation for backpropagation.
• Weight Initialization: Randomized within [-0.5, 0.5] to prevent symmetry issues.
• Training: Combines forward propagation, backpropagation, and gradient descent to minimize classification error. Stops when the average error per epoch drops below a certain threshold.

3. Results

• Training results show a rapidly decreasing error, stabilizing as the model optimizes for accurate digit recognition.

---

🛠 Usage

1. Draw the Digit:

   • Use the interactive canvas on the application’s main page to draw any digit between 0–9.

2. Submit for Prediction:

   • Click the "Predict" button after completing your drawing.

3. View Results:

   • The system will process your input and display the predicted digit.

4. Clear and Retry:

   • Use the "Clear" button to erase the canvas and draw a new digit for prediction.

---

📦 Dependencies

The project relies on the following dependencies:

• Django: Web framework for the backend.
• NumPy: For numerical processing.
• TensorFlow: Used only for the MNIST dataset, not for a pre-trained model.
• Pickle: Built-in python library for saving and loading the model.

---

🤝 Contributors

• Usman Faizyab Khan
• Muhammad Owais
• Muhammad Zunain
• Zuhaib Noor

---

Enjoy exploring the Digit Recognition System! Contributions and feedback are welcome. 😊