Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Online Handwritten Signature Verification Based on Temporal-Spatial Graph Attention Transformer

Hai-jie Yuan, Heng Zhang, Fei Yin

TL;DR

The paper addresses the challenge of robust online signature verification by modeling signatures as temporal-spatial graphs and learning their dynamics with a Graph Attention Transformer enhanced by dual graph representations and GRU temporal modeling. The Temporal-Spatial Graph Attention Transformer (TS-GATR) fuses a DGATR module for local and global spatial relationships with a GRU branch for long-range temporal dependencies, using DTW-based alignment and an Author-Invariant Triplet Loss to improve generalization across authors and forgeries. Key contributions include the DGATR dual-graph fusion, a temporal-spatial collaborative representation learning framework, and extensive ablations across two large benchmarks (MSDS and DeepSignDB) demonstrating state-of-the-art performance, including cross-session settings and diverse input modalities. The work advances practical OSV by delivering superior accuracy, template-diversity robustness, and a clear path toward real-time deployment and future extensions such as self-supervised learning and multimodal biometric fusion.

Abstract

Handwritten signature verification is a crucial aspect of identity authentication, with applications in various domains such as finance and e-commerce. However, achieving high accuracy in signature verification remains challenging due to intra-user variability and the risk of forgery. This paper introduces a novel approach for dynamic signature verification: the Temporal-Spatial Graph Attention Transformer (TS-GATR). TS-GATR combines the Graph Attention Network (GAT) and the Gated Recurrent Unit (GRU) to model both spatial and temporal dependencies in signature data. TS-GATR enhances verification performance by representing signatures as graphs, where each node captures dynamic features (e.g. position, velocity, pressure), and by using attention mechanisms to model their complex relationships. The proposed method further employs a Dual-Graph Attention Transformer (DGATR) module, which utilizes k-step and k-nearest neighbor adjacency graphs to model local and global spatial features, respectively. To capture long-term temporal dependencies, the model integrates GRU, thereby enhancing its ability to learn dynamic features during signature verification. Comprehensive experiments conducted on benchmark datasets such as MSDS and DeepSignDB show that TS-GATR surpasses current state-of-the-art approaches, consistently achieving lower Equal Error Rates (EER) across various scenarios.

Online Handwritten Signature Verification Based on Temporal-Spatial Graph Attention Transformer

TL;DR

The paper addresses the challenge of robust online signature verification by modeling signatures as temporal-spatial graphs and learning their dynamics with a Graph Attention Transformer enhanced by dual graph representations and GRU temporal modeling. The Temporal-Spatial Graph Attention Transformer (TS-GATR) fuses a DGATR module for local and global spatial relationships with a GRU branch for long-range temporal dependencies, using DTW-based alignment and an Author-Invariant Triplet Loss to improve generalization across authors and forgeries. Key contributions include the DGATR dual-graph fusion, a temporal-spatial collaborative representation learning framework, and extensive ablations across two large benchmarks (MSDS and DeepSignDB) demonstrating state-of-the-art performance, including cross-session settings and diverse input modalities. The work advances practical OSV by delivering superior accuracy, template-diversity robustness, and a clear path toward real-time deployment and future extensions such as self-supervised learning and multimodal biometric fusion.

Abstract

Handwritten signature verification is a crucial aspect of identity authentication, with applications in various domains such as finance and e-commerce. However, achieving high accuracy in signature verification remains challenging due to intra-user variability and the risk of forgery. This paper introduces a novel approach for dynamic signature verification: the Temporal-Spatial Graph Attention Transformer (TS-GATR). TS-GATR combines the Graph Attention Network (GAT) and the Gated Recurrent Unit (GRU) to model both spatial and temporal dependencies in signature data. TS-GATR enhances verification performance by representing signatures as graphs, where each node captures dynamic features (e.g. position, velocity, pressure), and by using attention mechanisms to model their complex relationships. The proposed method further employs a Dual-Graph Attention Transformer (DGATR) module, which utilizes k-step and k-nearest neighbor adjacency graphs to model local and global spatial features, respectively. To capture long-term temporal dependencies, the model integrates GRU, thereby enhancing its ability to learn dynamic features during signature verification. Comprehensive experiments conducted on benchmark datasets such as MSDS and DeepSignDB show that TS-GATR surpasses current state-of-the-art approaches, consistently achieving lower Equal Error Rates (EER) across various scenarios.
Paper Structure (20 sections, 17 equations, 2 figures, 8 tables)

This paper contains 20 sections, 17 equations, 2 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Method
  4. Symbol Description
  5. Graph Attention Layer
  6. Graph Structure Representation
  7. Dual-Graph Attention Transformer
  8. Temporal-spatial Collaborative Representation Learning
  9. Time-Functions Extraction
  10. Dynamic Time Warping
  11. Author-Invariant Triplet Loss
  12. Experiments
  13. Dataset Overview
  14. Experimental Setup
  15. Validation of Temporal-Spatial Modeling Components
  16. ...and 5 more sections

Figures (2)

  • Figure 1: The overall pipeline of the proposed TS-GATR model combines DGATR with GRU to fuse temporal and spatial features.
  • Figure 2: Overview of the Graph Attention Layer (GAL). The architecture integrates static topological constraints with dynamic attention propagation. The design preserves graph topology while enabling adaptive relationship modeling. B, L and d denote batch size, node count, and feature dimension, respectively.