You can deploy this model and start earning money today!

LasinduVIduranga/
sinhala-character-recognition-ml

Name: sinhala-character-recognition-ml
Availability: InStock

mit

Views

No views yet

language: si license: openmdw library_name: scikit-learn tags:

sinhala
character-recognition
knn
ocr
sinhala-nlp datasets:
sathiralamal/sinhala-letter-454 metrics:
accuracy pipeline_tag: image-classification

🇱🇰 Handwritten Sinhala Character Recognition Model

A K-Nearest Neighbors (KNN) classifier trained to recognize 454 Sinhala characters, including base characters, vowel modifications, and combined forms. This model is designed for handwritten Sinhala character recognition.

📊 Model Details

Property	Value
Model Type	K-Nearest Neighbors (KNN)
k Value	3 (default)
Classes	454 Sinhala characters
Input Shape	80×80 pixels (grayscale)
Feature Vector	6400 (flattened image)
Accuracy	~60%
Framework	Scikit-learn
Serialization	Joblib (.sav)

Dataset

This model was trained on the Sinhala Letter and Modifications Dataset.

Property	Value
Source	Kaggle - Sinhala Letter 454
Author	Sathira L. Amal
Classes	454 Sinhala characters
Image Size	80×80 pixels (grayscale)
Total Images	~45,000

Citation

bibtex

@dataset{sathiralamal_sinhala_2020,
  author       = {Sathira L. Amal},
  title        = {Sinhala Letter and Modifications},
  year         = {2020},
  publisher    = {Kaggle},
  url          = {https://www.kaggle.com/datasets/sathiralamal/sinhala-letter-454}
}

Usage

Installation

pip install joblib numpy opencv-python Pillow scikit-learn

Load and Predict

python

import joblib
import numpy as np
import cv2

# Load the model
model = joblib.load('sinahala-character-knn.sav')

# Load and preprocess your image
def preprocess_image(image_path):
    img = cv2.imread(image_path, 0)  # Grayscale
    img = cv2.resize(img, (80, 80))   # Resize to 80x80
    img = img.flatten()               # Flatten to 6400 features
    return img.reshape(1, -1)

# Make prediction
image_path = 'path/to/your/sinhala_character.jpg'
features = preprocess_image(image_path)
prediction = model.predict(features)
print(f'Predicted class: {prediction[0]}')

With Label Dictionary

python

label_dict = {
    1: 'අ', 2: 'ආ', 3: 'ඇ', 4: 'ඈ', 5: 'ඉ', 
    6: 'ඊ', 7: 'උ', 8: 'එ', 9: 'ඒ', 10: 'ඔ',
    11: 'ඕ', 12: 'ක', 13: 'කා', 14: 'කැ', 15: 'කෑ',
    16: 'කි', 17: 'කී', 18: 'කු', 19: 'කූ', 20: 'ක්',
    # ... full 454 mapping
}

# Map prediction to character
char = label_dict.get(prediction[0], 'Unknown')
print(f'Predicted character: {char}')

File Structure

sinahala-character-knn/
├── sinahala-character-knn.sav      # Trained KNN model (436 MB)
├── README.md                        # Model card
└── LICENSE                          # License file

Use Cases

Handwritten Sinhala character recognition
OCR systems for Sinhala text
Educational tools for learning Sinhala
Digitization of Sinhala manuscripts
Research in South Asian language processing

Performance Metrics

Metric	Value
Training Accuracy	~65%
Testing Accuracy	~60%
Precision (macro avg)	~0.60
Recall (macro avg)	~0.59
F1-Score (macro avg)	~0.59

Model Limitations

Accuracy: 60% accuracy means 4 out of 10 predictions may be incorrect.
Handwriting Variability: Works best on similar handwriting styles to training data.
No GPU Support: KNN runs on CPU only (no GPU acceleration).
Memory Intensive: Stores all training samples for prediction.

Future Improvements

Improve accuracy with Convolutional Neural Networks (CNN)
Data augmentation for better generalization
Real-time webcam recognition
Web deployment using Flask/Streamlit
Mobile app support (Android/iOS)
Support for cursive handwriting

Related Resources

License

This model is licensed under the OpenMDW License - see the LICENSE file for details.

Acknowledgments

Dataset: Sathira L. Amal for the Sinhala Letter 454 dataset
Scikit-learn community
OpenCV community
Hugging Face for hosting

Contact

Lasindu Viduranga - wplvidu@gmail.com

Model Link: https://huggingface.co/LasinduViduranga/sinhala-character-knn

GitHub Link: https://github.com/LasinduViduranga/sinhala-character-recognition-ml

Support

If you find this model useful, please:

Star the GitHub repository
Like this model on Hugging Face
Cite the dataset if you use it in research

Made with ❤️ for the Sinhala language community

LasinduVIduranga/sinhala-character-recognition-ml