MNIST digit recognition using Convolutional Neural Networks

This project explores digit recognition using Convolutional Neural Networks (CNNs) with the MNIST dataset. It involves building and training models in TensorFlow/Keras to classify handwritten digits. It also includes a simple app (main.py) where you can handwrite a digit and see real time predictions from the saved model.

Project Overview

• Datasets: MNIST digits. Source: https://www.kaggle.com/datasets/hojjatk/mnist-dataset?resource=download
• Model: CNN.
• Techniques: Data Augmentation, Dropout, Early Stopping, LRS, Batch Normalization
• Goal: Classify handwritten digits into one of 10 categories i.e. 0-9.

Project Structure

📂 digit-recognition-MNIST
│
├── 📂 model/mnist_model.keras         #Saved model
├── main.py
├── character-reader.ipynb            # Main script
├── requirements.txt            # Project Dependencies
└── .gitignore

---

Installation and Setup

1. Set the destination folder
   Create or navigate to the folder where you want this project to be located:

2. Clone the repository:

   git clone https://github.com/P-Alakara/digit-recognition-MNIST.git

3. Navigate to the newly created folder.

    cd digit-recognition-MNIST

4. Create and activate a virtual environment (optional but recommended):

   python -m venv venv
   source venv/bin/activate  # venv\Scripts\activate (for windows systems)

5. Install dependencies:

   pip install -r requirements.txt

6. Run the app

   python main.py

---