Conclusion
- Writer-dependent models (Random Forest) show substantially better performance, with a 12.75% increase in accuracy, much lower error rates, and overall more robust verification compared to the writer-independent approach (Siamese Network).
- TimeSeries Model: This Timeseries model performed well but had limitations in capturing the complexity of certain forged signatures.
- DenseNet Model: Leveraging additional feature extraction through DenseNet improved accuracy and reduced error rates, particularly for complex signatures.
Further Work
- Involve combining other deep learning techniques or experimenting with hybrid models to further improve accuracy and robustness against sophisticated forgeries.
- Evaluate these models for their performance under adversarial attack scenarios and also evaluate if they are biased.
Acknowledgement: We sincerely thank the Bucknell University College of Engineering for their sponsorship.
Related work
- Previous studies have demonstrated promise in CNN-based deep learning method for online signature verification
- Traditional machine learning models that relied on handcrafted features, especially when trained on large datasets have also showed promise.
- Research have emphasized the importance of time-series features such as velocity and acceleration in capturing the behavioral traits of a user’s signature, leading to more accurate online signature verification models.
- These studies however, have been conducted on one datasets
- In this work we evaluated these methods on multiple datasets to benchmark the performance of static and dynamic features.
References
- [1] Xie, L., Wu, Z., Zhang, X., Li, Y., & Wang, X. (2022). Writer-independent online signature verification based on 2D representation of time series data using triplet supervised network. Measurement, 197, 111312.
- [2] Xia, X., Song, X., Luan, F., Zheng, J., Chen, Z., & Ma, X. (2018). Discriminative feature selection for on-line signature verification. Pattern Recognition, 74, 422-433.
- [3] Taherzadeh, G., Karimi, R., Ghobadi, A., & Beh, H. M. (2011). Evaluation of online signature verification features. In 13th International Conference on Advanced Communication Technology (ICACT2011) (pp. 772–777).
Background and motivation
- Online signature verification is essential for security in applications such as banking and legal documentation, where authenticity is paramount.
- Static features, such as shape and structure, and dynamic features, like pressure and speed, both offer valuable insights, but their comparative effectiveness requires further exploration.
- We evaluate two online signature verification methods on two widely used datasets.
Benchmarking Static and Dynamic Features for Online Signature Verification
Duy Le, Nhi Cao, Rajesh Kumar | Computer Science, Bucknell University, PA, USA
Experimental details
- Offline signature: Typically handwritten signatures on paper using a pen or
pencil. Then a system will read the signature scan and verify.
- Online signature: Are created using a stylus or finger or digital pen on a
digital surface, such as a tablet or touchscreen.
- Datasets:
- SVC 2004: Capture signatures in a more controlled environment on digital
signature tablets.
- MOBISIG: signatures captured under different conditions using mobile devices
such as smartphones and tablets.
- Preprocessing data for both methods:
- Offline signatures: Generated images from x, y coordinates, followed by data
augmentation (rotation, scaling, zoom, shift) to improve robustness and
prevent overfitting.
- Online Signatures: Extracted key features from raw time-series data (x, y coordinates, timestamp, pressure, velocity, acceleration). The top 50 features were selected using a Random Forest Classifier for enhanced model performance.
- Verification pipeline:
- Time Series Model: Takes the raw time-series data of a signature and outputs a binary classification (genuine or forgery) based on trained classifiers such as MLP, KNN, RF, etc.
- DenseNet Model: The DenseNet processes an image-like representation of the signature’s trajectory and outputs feature embeddings, which are then combined with time-series features to improve the classification.
Table 1: Detail of each dataset
| | |
| | |
Real signature sample for each user | | |
Fake signature samples for each user | | |
Fig 4. Heatmap showing results for Timeseries method on SVC 2004 dataset
Result highlights
- Time Series Model (RF): Achieved 95.75% accuracy with 0.015 FRR and 0.0275 FAR for SVC 2004 dataset.
- DenseNet (RF): Resulted in accuracy of 89% with reduced FAR for SVC 2004 dataset.
Fig 1. A sample online signature from the MOBISIG dataset
Fig 5. Heatmap showing Timeseies results for MOBISIG dataset
Fig 6. Heatmap showing DenseNet results for SVC 2004 dataset
Fig 7. Heatmap showing DenseNet results for MOBISIG dataset
Poster Print Size:
This poster template is 36 high by 48 wide. It can be used to print any poster with a 3:4 aspect ratio.
Placeholders:
The various elements included in this poster are ones we often see in medical, research, and scientific posters. Feel free to edit, move, add, and delete items, or change the layout to suit your needs. Always check with your conference organizer for specific requirements.
Image Quality:
You can place digital photos or logo art in your poster file by selecting the Insert, Picture command, or by using standard copy & paste. For best results, all graphic elements should be at least 150-200 pixels per inch in their final printed size. For instance, a 1600 x 1200 pixel photo will usually look fine up to 8-10 wide on your printed poster.
To preview the print quality of images, select a magnification of 100% when previewing your poster. This will give you a good idea of what it will look like in print. If you are laying out a large poster and using half-scale dimensions, be sure to preview your graphics at 200% to see them at their final printed size.
Please note that graphics from websites (such as the logo on your hospital's or university's home page) will only be 72dpi and not suitable for printing.
�[This sidebar area does not print.]
Change Color Theme:
This template is designed to use the built-in color themes in the newer versions of PowerPoint.
To change the color theme, select the Design tab, then select the Colors drop-down list.
The default color theme for this template is Office, so you can always return to that after trying some of the alternatives.
Printing Your Poster:
Once your poster file is ready, visit www.genigraphics.com to order a high-quality, affordable poster print. Every order receives a free design review and we can deliver as fast as next business day within the US and Canada.
Genigraphics® has been producing output from PowerPoint® longer than anyone in the industry; dating back to when we helped Microsoft® design the PowerPoint® software.
US and Canada: 1-800-790-4001�Email: info@genigraphics.com
�[This sidebar area does not print.]