Indic-OCR: Multi-Language OCR System for Indian Regional Scripts

A machine learning-based OCR solution that can detect, classify, and accurately recognize text from multiple Indian regional scripts and convert it into editable Unicode text.

📋 Project Information

• Title: Multi-Language OCR System for Indian Regional Scripts (Indic-OCR)
• Author: Dhananjayan H
• Roll No: AA.SC.P2MCA24070151
• Course: MCA Minor Project (21CSA697A)

🎯 Objectives

1. Develop a multi-language OCR system capable of identifying the script and extracting text from handwritten or printed Indian regional languages
2. Convert extracted text into Unicode digital text
3. Support applications like digitization of academic notes, historical records, government documents, and accessibility enhancement

✨ Features

• Multi-Script Support: Devanagari (Hindi), Malayalam, Tamil
• Script Detection: Automatic identification of the input script
• Image Preprocessing: Advanced preprocessing for improved accuracy
• Deep Learning OCR: State-of-the-art recognition using PaddleOCR/TrOCR
• Web Interface: User-friendly Streamlit-based interface
• Evaluation Tools: CER/WER metrics for performance assessment

🏗️ Project Structure

Indic-OCR/
├── app/
│   └── streamlit_app.py       # Web interface
├── config/
│   └── config.yaml            # Configuration file
├── data/
│   ├── raw/                   # Raw dataset
│   ├── processed/             # Processed images
│   ├── train/                 # Training set
│   ├── val/                   # Validation set
│   └── test/                  # Test set
├── models/                    # Saved models
├── notebooks/                 # Jupyter notebooks
├── outputs/                   # OCR outputs
├── src/
│   ├── __init__.py
│   ├── preprocessing.py       # Image preprocessing
│   ├── script_classifier.py   # Script identification model
│   ├── ocr_engine.py          # OCR recognition
│   ├── dataset.py             # Dataset utilities
│   ├── evaluation.py          # Evaluation metrics
│   └── utils.py               # Utility functions
├── tests/                     # Unit tests
├── requirements.txt           # Dependencies
├── train.py                   # Training script
├── inference.py               # Inference script
└── README.md                  # Documentation

🚀 Installation

Prerequisites

• Python 3.8 or higher
• pip package manager
• GPU (optional, for faster training)

Setup

1. Clone the repository:

cd "MCA Project/Indic-OCR"

2. Create a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. For GPU support (optional):

pip install paddlepaddle-gpu  # For CUDA 11.x
# or
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118

📊 Dataset

Supported Datasets

1. Public Datasets:

   • Devanagari Handwritten Character Dataset
   • Tamil Handwritten Dataset
   • BHaratWrites Dataset

2. Self-collected Data:

   • Scanned documents
   • Handwritten notes
   • Printed materials

Data Preparation

python -c "from src.dataset import DatasetManager; dm = DatasetManager(); dm.split_dataset()"

🎓 Training

Train Script Classifier

python train.py --task script_classifier --epochs 50 --batch_size 32

Train OCR Model (Fine-tuning)

python train.py --task ocr --model paddleocr --language hindi

🔮 Inference

Command Line

python inference.py --image path/to/image.png --output results.txt

Python API

from src import ImagePreprocessor, ScriptClassifier, OCREngine

# Initialize components
preprocessor = ImagePreprocessor()
classifier = ScriptClassifier()
ocr = OCREngine()

# Process image
image = preprocessor.preprocess("document.png")
script, confidence = classifier.predict(image)
result = ocr.recognize(image, language=script)

print(f"Detected Script: {script}")
print(f"Extracted Text: {result['text']}")

🌐 Web Interface

Launch the Streamlit web application:

cd Indic-OCR
streamlit run app/streamlit_app.py

Access the interface at http://localhost:8501

📈 Evaluation

Run Evaluation

python -c "from src.evaluation import evaluate_ocr_results; evaluate_ocr_results(predictions, ground_truths)"

Metrics

• CER (Character Error Rate): Measures character-level accuracy
• WER (Word Error Rate): Measures word-level accuracy
• Accuracy: Percentage of correctly recognized samples

📅 Timeline & Milestones

Week	Milestone
1	Literature review & requirements analysis
2	Dataset collection & preprocessing
3	Script identification model training
4	OCR engine integration & testing
5	UI development & deployment
6	Accuracy evaluation & improvements
7	Final testing, documentation & presentation

🛠️ Tools & Technologies

Category	Tools
Language	Python
Libraries	OpenCV, PaddleOCR/TrOCR, TensorFlow/PyTorch
IDE	VS Code, Jupyter Notebook
Interface	Streamlit
Version Control	GitHub
Hardware	Laptop (8GB+ RAM) + Google Colab GPU

📚 Learning Outcomes

• ML & DL: Model training, image classification, evaluation
• Computer Vision: Preprocessing, thresholding, deskewing, feature extraction
• OCR Systems: Text detection, recognition, and Unicode conversion
• Research Methodology: Dataset preparation, benchmarking, literature review
• Web Development: Creating a functional front-end for OCR usage

📖 References

1. PaddleOCR Documentation: https://github.com/PaddlePaddle/PaddleOCR
2. TrOCR Paper: https://arxiv.org/abs/2109.10282
3. OpenCV Documentation: https://docs.opencv.org/
4. Streamlit Documentation: https://docs.streamlit.io/

📄 License

This project is developed purely for academic and research-related purposes.

👤 Author

Dhananjayan H
Roll No: AA.SC.P2MCA24070151
Department of Computer Science

---

Last Updated: November 2024