Related Experiment Video
Updated: May 6, 2026
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
9.9K
Contactless biometric verification from in-air signatures using deep siamese networks
Serkan Salturk1, Taha Emre Pamukcu2, Nihan Kahraman2
1Department of Informatics, Yildiz Technical University, 34320, Istanbul, Turkey. ssalturk@yildiz.edu.tr.
Scientific Reports
|January 3, 2026
Summary
In-air signature verification uses deep learning to authenticate individuals via contactless 3D gestures. This novel biometric system achieves 85% accuracy, proving effective for secure remote authentication.
Area of Science:
- Biometrics
- Computer Science
- Machine Learning
Background:
- In-air signatures offer contactless, hygienic, and remote authentication.
- Traditional methods lack flexibility; in-air capture uses 3D spatial data.
Purpose of the Study:
- Develop a deep learning model for in-air signature verification.
- Evaluate the model's generalization across users.
Main Methods:
- Collected in-air signature data from 25 participants.
- Utilized a Siamese Neural Network with Bidirectional LSTM and contrastive loss.
- Employed a Leave Two Sample Out (LTSO) cross-validation protocol.
Main Results:
- Achieved 85% average accuracy, 85% F1-score, and 91% recall.
- Demonstrated effectiveness even with limited training data.
- Showcased strong generalization capability across users.
Conclusions:
- In-air signatures are a viable contactless biometric modality.
- Siamese neural networks effectively learn person-specific motion patterns for verification.
