rakshitdabral/Signature-Verification-Dataset

Multilingual Signature Verification Dataset

Dataset Summary

The Multilingual Signature Verification Dataset is a curated collection of handwritten signatures designed for offline signature verification and related computer vision tasks.

The dataset contains more than 7,000 signature images spanning three major writing systems:

• —Hindi
• —Bengali
• —English

The English portion includes samples from the well-known CEDAR Signature Dataset, while additional Hindi and Bengali signatures have been curated to support multilingual signature verification research.

The dataset has been preprocessed and organized into ready-to-use train,test and valid splits for machine learning workflows.

Supported Tasks

This dataset can be used for:

• —Offline Signature Verification
• —Signature Classification
• —Signature Recognition
• —Siamese Network Training
• —Contrastive Learning
• —Metric Learning
• —Image Embedding Models
• —Document AI Research
• —Handwritten Biometrics

Dataset Structure

dataset/
├── train/
│   ├── person_0001/
│   ├── person_0002/
│   ├── ...
│
└── test/
    ├── person_0005/
    ├── person_0021/
    ├── ...

Each class corresponds to a unique signer identity.

Dataset Statistics

Languages Included

English

The English subset contains signatures derived from the CEDAR Signature Dataset, one of the most widely used benchmark datasets for offline signature verification research.

Hindi

Contains handwritten signatures written in the Devanagari script.

Bengali

Contains handwritten signatures written in the Bengali script.

The multilingual composition enables research on language-independent and script-independent signature verification models.

Loading the Dataset

Using the 🤗 Datasets library:

from datasets import load_dataset

dataset = load_dataset("rakshitdabral/Signature-Verification-Dataset")

Example:

train = dataset["train"]
test = dataset["test"]

print(train[0])

Example Training Applications

This dataset is suitable for training:

• —ResNet
• —ConvNeXt
• —Vision Transformer (ViT)
• —EfficientNet
• —MobileNet
• —Siamese Networks
• —Triplet Networks
• —ArcFace-based Models
• —Contrastive Learning Pipelines

Recommended Evaluation Metrics

• —Accuracy
• —Precision
• —Recall
• —F1 Score
• —ROC-AUC
• —Equal Error Rate (EER)
• —False Acceptance Rate (FAR)
• —False Rejection Rate (FRR)

Intended Use

This dataset is intended for:

• —Academic research
• —Signature verification systems
• —Identity verification research
• —Document processing
• —Machine learning benchmarking
• —Deep learning research
• —Educational purposes

Limitations

• —The dataset focuses on offline handwritten signatures only.
• —It is not intended for online signature dynamics (pressure, speed, stroke order).
• —Performance may vary across scripts depending on model architecture and training strategy.

Citation

If you use this dataset in your research, please cite this repository.

@dataset{signature_verification_dataset,
  title={Multilingual Signature Verification Dataset},
  author={Rakshit Dabral},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/rakshitdabral/Signature-Verification-Dataset}
}

License

This dataset is released under the MIT License.

Please ensure compliance with the licensing terms of any original source datasets (such as the CEDAR Signature Dataset) when using or redistributing derived data.

Acknowledgements

• —CEDAR Signature Dataset
• —Hugging Face Datasets
• —Open-source computer vision community

Contact

For questions, issues, or contributions, please open an issue on the Hugging Face repository or contact the dataset maintainer.