Online Signature Verification based on the Lagrange formulation with 2D and 3D robotic models

2025-03-01

Pattern Recognition

AI Summary

Online signature verification typically relies on shallow temporal coordinate and pressure features, failing to capture the underlying biomechanical motion mechanisms of signers. To address this, we propose the first physics-consistent modeling framework grounded in Lagrangian mechanics, unifying 2D handwriting dynamics with 3D robotic arm kinematics to yield interpretable dynamic representations of signing behavior. We innovatively introduce the Lagrangian equation into biometric verification, designing both generative and discriminative temporal models under physical constraints, and integrating physics-guided neural embeddings with online time-series alignment. Evaluated on SigComp2024 and DeepSignDB benchmarks, our method achieves an 18.7% reduction in equal error rate (EER), significantly enhancing robustness against replay and imitation attacks. With inference latency under 35 ms, it satisfies real-time verification requirements.

Technology Category

• Computer Vision: Low Level & Physics-based Vision
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior
• Intelligent Robots: Manipulation

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Problem

• Enhance online signature verification using robotic models.
• Infer arm pose and dynamics from digitizer data.
• Integrate kinematic and dynamic features for deep learning.

Innovation

• Lagrangian formulation for signature dynamics
• 2D and 3D robotic arm models
• Deep learning integration for verification