proadhikary/dzongkha-digits

Dzongkha Handwritten Digit Dataset

This dataset was created to address the lack of publicly available resources for handwritten recognition of the Dzongkha script—the national language of Bhutan. By offering a benchmark dataset, it aims to spur research in OCR, pattern recognition, and machine learning for underrepresented languages and scripts. A benchmark dataset of handwritten Dzongkha digit images, developed to support research in optical character recognition for a low-resource script.

---

Dataset Overview

• Authors: Tawmo, Prottay Kumar Adhikary, Pankaj Dadure, and Partha Pakray (National Institute of Technology Silchar) oai_citation:0‡Zenodo
• Publication Date: February 25, 2022 oai_citation:1‡Zenodo
• License: Creative Commons Attribution 4.0 International (CC-BY-4.0) oai_citation:2‡Zenodo
• Format: JPG images collected via Google Jamboard
• Size & Classes:
  • Total of 1,000 images
  • Data from 100 participants (diverse in background)
  • Ten classes: Dzongkha digits 0–9 (༠–༩) oai_citation:3‡Zenodo oai_citation:4‡NIAID Data Ecosystem Discovery Portal

---

Associated Publication

The dataset supports the work titled: Dzongkha Handwritten Digit Recognition using Machine Learning Techniques
Prottay Kumar Adhikary, Pankaj Dadure, Pradipta Saha, Partha Pakray — Procedia Computer Science, Volume 218, January 2023 oai_citation:5‡scholar.google.co.jp oai_citation:6‡proadhikary.github.io.

This study evaluates various classical machine learning algorithms—including SVM, K-NN, and Decision Tree—for recognizing Dzongkha handwritten digits, achieving up to 98.29% accuracy with a Support Vector Machine oai_citation:7‡proadhikary.github.io.

---

Usage Examples

from datasets import load_dataset

# Replace with actual path or transfer the dataset to Hugging Face
dataset = load_dataset("proadhikary/dzongkha-digits")

# Inspect the dataset
print(dataset)
# Example processing:
# image = dataset["train"][0]["image"]
# label = dataset["train"][0]["label"]

Typical workflows include:

• Preprocessing: resizing, normalization, grayscale conversion, augmentation
• Modeling: training classical ML models (SVM, KNN) or simple CNNs
• Evaluation: accuracy, confusion matrices, per-digit error analysis

Recommended Citation

If you use this dataset in your work, please cite both the dataset and the associated paper:

@dataset{tawmo_2022_6271560,
  author = {Tawmo and Prottay Kumar Adhikary and Pankaj Dadure and Partha Pakray},
  title = {Dzongkha Handwritten Digit Dataset},
  year = {2022},
  publisher = {Zenodo},
  doi = {10.5281/zenodo.6271560},
  url = {https://doi.org/10.5281/zenodo.6271560}
}

@article{adhikary2023dzongkha,
  author = {Prottay Kumar Adhikary and Pankaj Dadure and Pradipta Saha and Partha Pakray},
  title = {Dzongkha Handwritten Digit Recognition using Machine Learning Techniques},
  journal = {Procedia Computer Science},
  volume = {218},
  pages = {...},
  year = {2023}
}