✏️ Handwritten Digits Recognition 🔢

📝 Overview

This project implements a neural network model to recognize handwritten digits using the MNIST dataset. The model is built with TensorFlow and achieves high accuracy in classifying digits from 0-9.

🔍 Features

• 🧠 Neural network with multiple dense layers
• 📊 Training and evaluation on the MNIST dataset
• 💾 Model saving and loading capabilities
• 🖼️ Custom image prediction functionality
• 📉 Visualization of results using matplotlib

🛠️ Technologies Used

• Python
• TensorFlow
• NumPy
• OpenCV
• Matplotlib

🏗️ Model Architecture

The neural network consists of:

• Input layer (flattened 28×28 pixels)
• Two hidden layers with 128 neurons each and ReLU activation
• Output layer with 10 neurons and softmax activation

🚀 How to Use

1. Install the required dependencies:

pip install tensorflow opencv-python numpy matplotlib

2. Set train_new_model = True to train the model or False to use a pre-trained model.

3. Place your own handwritten digit images in the digits/ folder with naming format digit1.png, digit2.png, etc.

4. Run the script:

python handwritten_digits_recognition.py

📋 Results

The model achieves approximately 97-98% accuracy on the MNIST test set after just 3 epochs of training.

🎓 Academic Context

This project was developed as part of the IT3140 Soft Computing course during my undergraduate studies at Manipal University Jaipur.

📊 Future Improvements

• Implement data augmentation for better generalization
• Try different model architectures (CNN, etc.)
• Create a web interface for real-time digit recognition
• Extend to recognize characters beyond digits

🔗 References

• MNIST Dataset
• TensorFlow Documentation