🏛️ Brahmi Script Recognition & Devanagari Transliteration

Deep Learning-based Computer Vision System for Ancient Brahmi Script Recognition and Devanagari Transliteration using MobileNetV2 Transfer Learning.

---

📖 Overview

Brahmi is one of the oldest writing systems of the Indian subcontinent and is considered the ancestor of many modern Indian scripts. Recognizing and transliterating Brahmi inscriptions manually requires expert knowledge and is often time-consuming.

This project presents an end-to-end Deep Learning pipeline capable of recognizing Brahmi script characters from images and transliterating them into their corresponding Devanagari characters. The system is powered by MobileNetV2 Transfer Learning, a custom-built dataset, image preprocessing techniques, and a Streamlit web application for real-time inference.

---

✨ Features

• 📜 Recognizes ancient Brahmi characters
• 🔤 Converts Brahmi characters into Devanagari
• 🧠 MobileNetV2 Transfer Learning
• 🖼 Automatic image preprocessing
• 📈 Confidence score prediction
• ⚖ Balanced dataset generation
• 🚀 Streamlit-based web application
• 💾 JSON-based character mapping
• 📊 Training visualization and evaluation

---

🏗️ System Architecture

---

📂 Dataset

The dataset was prepared using a custom preprocessing and augmentation pipeline.

Property	Value
Total Classes	416
Total Images	6240
Images per Class	15
Image Size	224 × 224

Dataset Preparation Pipeline

Original Dataset
        │
        ▼
Dataset Cleaning
        │
        ▼
Dataset Balancing
        │
        ▼
Image Augmentation
        │
        ▼
Image Preprocessing
        │
        ▼
Processed Dataset

---

🧠 Model Architecture

The recognition model uses Transfer Learning with MobileNetV2.

Input Image
      │
      ▼
Resize (224×224)
      │
      ▼
Normalization
      │
      ▼
MobileNetV2 Feature Extractor
      │
      ▼
Global Average Pooling
      │
      ▼
Dense Layer (ReLU)
      │
      ▼
Dropout
      │
      ▼
Softmax (416 Classes)
      │
      ▼
Predicted Character
      │
      ▼
JSON Character Mapping
      │
      ▼
Devanagari Output

---

📊 Training Results

Metric	Value
Model	MobileNetV2
Optimizer	Adam
Epochs	30
Validation Accuracy	72.44%
Validation Loss	1.0711

Accuracy Curve

Loss Curve

---

🖥️ Application Demo

Home Page

---

Upload Image

---

Prediction Result

---

📁 Project Structure

Brahmi-Transliteration
│
├── assets/
├── configs/
├── fonts/
├── mapping/
├── model/
├── scripts/
│
├── app.py
├── README.md
├── requirements.txt
├── LICENSE
└── pyproject.toml

---

⚙️ Installation

Clone the repository

git clone https://github.com/krishnikbarman/Brahmi-Transliteration.git
cd Brahmi-Transliteration

Install dependencies

pip install -r requirements.txt

Run the application

streamlit run app.py

---

🚀 Usage

1. Launch the Streamlit application.
2. Upload a Brahmi character image.
3. The image is automatically preprocessed.
4. The trained MobileNetV2 model predicts the character.
5. The corresponding Devanagari character and confidence score are displayed.

---

🛠️ Technologies Used

• Python
• TensorFlow
• Keras
• MobileNetV2
• OpenCV
• NumPy
• Pillow
• Matplotlib
• Streamlit
• JSON

---

📌 Applications

• Ancient Script Recognition
• Historical Document Digitization
• Cultural Heritage Preservation
• Educational Learning Tools
• OCR Research
• Computer Vision Research

---

🔮 Future Improvements

• Support for handwritten Brahmi characters
• Sentence-level transliteration
• Vision Transformer (ViT) based models
• Domain adaptation for user-drawn inputs
• Mobile application deployment
• Ancient inscription recognition from photographs

---

👨‍💻 Author

Krishnik Barman

B.Tech in Computer Science & Engineering

Areas of Interest

• Artificial Intelligence
• Deep Learning
• Computer Vision
• OCR Systems
• Ancient Script Digitization

---

📄 License

This project is released under the MIT License.

---

⭐ Support

If you found this project useful, consider giving it a ⭐ on GitHub.