MS-SigNet

MultiScale Signature feature learning Network for offline handwritten signature verification

1. The proposed MS-SigNet can capture and integrate global and regional information from various spatial scales to generate discriminative features.
2. The proposed co-tuplet loss can learn the distance metric for handwritten signature verification. The loss aims to transform input features into a feature space where genuine signatures from the same writer are close to each other while corresponding forgeries are far away from genuine ones.

Experimental Results

Performance comparison between different combinations of models and losses (evaluation metrics in %)

Implementation

• Python ≥ 3.8
• PyTorch framework
• NVIDIA GPUs are needed for both training and testing

Citation

If you use MS-SigNet or co-tuplet loss in your research, please cite our work:
F.-H. Huang and H.-M. Lu. Multiscale Feature Learning Using Co-Tuplet Loss for Offline Handwritten Signature Verification. arXiv preprint arXiv:2308.00428, 2023.

@misc{huang2023multiscale,
      title = {Multiscale Feature Learning Using Co-Tuplet Loss for Offline Handwritten Signature Verification}, 
      author = {Fu-Hsien Huang and Hsin-Min Lu},
      year = {2023},
      eprint = {2308.00428},
      archivePrefix = {arXiv},
      primaryClass = {cs.CV}
}

License

This code is distributed under MIT license (refer to the LICENSE file for details).