Welcome to the Persian Alphabet Recognition project! This repository implements a robust system for classifying 43 Persian alphabet characters using two neural network models: a fully connected network and a convolutional neural network (CNN). 🚀

This project processes image datasets, trains two neural network models, evaluates their performance, and visualizes results. It includes data preprocessing, model training, ROC/AUC analysis, and real-world image classification. 🖼️🔍

• Data Preprocessing 🛠️: Loads and preprocesses images (64x64, normalized, inverted) using a custom DataLoader class.
• Model 1 🧠: Fully connected neural network with dense layers (2048 to 64 units, ReLU, softmax).
• Model 2 🌐: CNN with Conv2D, BatchNormalization, MaxPooling, and Dropout layers.
• Training 📈: Both models trained for 20 epochs using Adam optimizer and sparse categorical crossentropy loss.
• Evaluation 📊: ROC curves, AUC scores, and accuracy/loss plots for performance analysis.
• Visualization 🎨: Displays sample images, predictions, and real-world test results using Matplotlib.
• Real-World Testing 🌍: Classifies preprocessed real-world images with both models.

• Python 3.x 🐍
• Keras
• NumPy
• Pandas
• Matplotlib
• Scikit-learn
• OpenCV

1. Clone the repository:

   git clone https://github.com/shahin-ro/persian-alphabet-recognition.git

2. Install dependencies:

   pip install -r requirements.txt

1. Update dataset paths (DATASET1, DATASET2, DATASET3, REAL_DATA) in alphabet_recognition.py.

2. Run the script:

   python alphabet_recognition.py

• Model 1 Test Accuracy ✅: Displayed in console output.
• Model 2 Test Accuracy ✅: Displayed in console output.
• Visualizations include:
  • Training/validation accuracy and loss plots 📈
  • ROC curves for each class 📉
  • Sample test images with true/predicted labels 🖼️
  • Real-world image predictions 🌍

• machine-learning
• deep-learning
• neural-network
• cnn
• image-classification
• persian-alphabet
• keras
• python
• computer-vision
• data-science

This project is licensed under the MIT License. See the LICENSE file for details. 📝

Contributions are welcome! Feel free to open issues or submit pull requests to improve the project. 🤝

For questions or feedback, reach out via GitHub Issues or email at [email protected] 📬