✍️ Offline Handwritten Character Recognition Desktop App

A standalone desktop application that recognizes handwritten letters (A-Z, a-z) and digits (0-9) in real-time using a Convolutional Neural Network (CNN). Works completely offline with 93%+ accuracy.

🎥 Demo

📖 Table of Contents

• Getting Started
  • Step 1: Install Python 3.10.0 or higher version
  • Step 2: Clone the repository
  • Step 3: Install Required Libraries
  • Step 4: Train the Models
  • Step 5: Run the App

📂 Project Structure

The project should be organized with the following folder and file structure. Do not rename the data files.

deep-learning-handwritten-recognition/
│
├── app/
│   └── app.py
│
├── assets/
|   ├── 6.png
|   └── L.png
|    
├── training/
|   └── brain.py
|
├── handwriting_model.py
├── README.md
└── requirements.txt

🧰 System Requirements

Requirement	Details
Python Version	3.10 or above (Crucial for compatibility)
Operating System	Windows / Linux / macOS
Internet Access	Required only for initial library installation

🚀 Getting Started

Follow these steps to set up the project, train the models, and run the visualization application.

Step 1: Install Python 3.10.0

It is essential to use the specified Python version for library compatibility.

1. Download Python 3.10.0 from the official website: https://www.python.org/downloads/

2. During the installation process, ensure you check the box for Add Python to PATH.

3. Verify the installation by opening a terminal or command prompt and running:

   python --version

   The output should confirm the version: Python 3.10.0.

Step 2: Clone the repository

```bash
git clone https://github.com/AreebaGhaffar/deep-learning-handwritten-recognition.git
cd deep-learning-handwritten-recognition
```

Step 3: Install Required Libraries

Ensure the requirements.txt file is present in the root directory.

1. Upgrade pip and install all required libraries:

   pip install --upgrade pip
   pip install -r requirements.txt

   Wait for all packages to install successfully.

Step 4: Train the Models

Both the digit and letter classification models must be trained before running the application. Training may take a few minutes depending on your CPU.

🔤 Train digit annd Letter (EMNIST) Model

Run the following command to train the CNN on EMNIST data:

python brain.py

This script will load the EMNIST dataset, train the classification model, and save the weights (e.g., as handwriting_model.h5).

Step 5: Run the App

Once model is trained and saved, you can launch the interactive visualization application.

python app.py

it will automatically open the application.

✨ Features

• 📝 Real-time Recognition: Draw any character and get instant predictions
• 🌐 Fully Offline: No internet connection required after setup
• 🔊 Text-to-Speech: Optional voice output for predictions
• 🎨 Clean GUI: Modern interface built with CustomTkinter
• 🧠 Custom CNN: Model trained from scratch on EMNIST dataset
• 🖱️ Easy to Use: Simple drawing canvas with clear controls

🏗️ Architecture

User Draws → Canvas → Preprocessing → CNN Model → Prediction → Display/Speak (Tkinter) (PIL/NumPy) (TensorFlow) (Tkinter/pyttsx3)