MNIST Neural Network Classifier

This project demonstrates how to train a neural network using TensorFlow on the MNIST dataset (handwritten digits), save the trained model, and use it to predict digits from new image files.

---

📌 Features

• Trains a simple yet effective deep neural network on the MNIST dataset.
• Saves the trained model to disk.
• Loads the saved model for future predictions.
• Predicts digit values from custom image inputs (.png files).
• Visualizes each processed image along with its prediction.

---

🧠 Model Architecture

• Input Layer: Flatten (28x28 pixels)
• Hidden Layers:
  • Dense Layer (128 units, ReLU)
  • Dense Layer (128 units, ReLU)
• Output Layer: Dense Layer (10 units, Softmax for digit classification 0–9)

---

🛠️ Requirements

Install the dependencies using pip:

pip install numpy opencv-python matplotlib tensorflow

---

📁 Project Structure

project/
├── digits/
│   ├── digit1.png
│   ├── digit2.png
│   └── ...
├── mnist_model.h5
└── main.py

---

🚀 How to Run

1. Train the Model (Optional) If you want to retrain the model from scratch, uncomment and run the training section in mnist_classifier.py:

(x_train, y_train), (x_test, y_test) = mnist.load_data()
model.fit(x_train, y_train, epochs=3)
model.save('mnist_model.h5')

2. Add Custom Digit Images Save grayscale images (28x28 pixels recommended) of digits to the digits/ folder.

Name them sequentially as digit1.png, digit2.png, etc.

3. Run Prediction

python main.py The script will load mnist_model.h5, preprocess each image, and display the predicted digit.

Predictions are shown with the corresponding image using matplotlib.

---

📷 Image Format Guidelines

Images should be in grayscale (cv2.IMREAD_GRAYSCALE is used).

Recommended to use white digits on black background.

Each image will be:

Resized to 28x28.

Inverted (black to white) to match MNIST data format.

Normalized to [0, 1] range.

---

📊 Sample Output

Prediction for image 1: 1
Prediction for image 2: 2

Displays each image along with its predicted digit.

---