[Paper Review] Improved Dynamic Time Warping (DTW) Approach for Online Signature Verification

Azhar Ahmad Jaini, Ghazali Sulong+1 more|arXiv (Cornell University)|Mar 26, 2019

Handwritten Text Recognition Techniques35 references4 citations

TL;DR

This paper proposes an enhanced Dynamic Time Warping (DTW) algorithm with a novel Lost Box Recovery mechanism to improve stroke mapping in online signature verification. By addressing misalignment in time-series signature data, the method achieves higher verification accuracy than conventional DTW, particularly in handling temporal distortions during real-time signature comparison on tablet devices.

ABSTRACT

Online signature verification is the process of verifying time series signature data which is generally obtained from the tablet-based device. Unlike offline signature images, the online signature image data consists of points that are arranged in a sequence of time. The aim of this research is to develop an improved approach to map the strokes in both test and reference signatures. Current methods make use of the Dynamic Time Warping (DTW) algorithm and its variant to segment them before comparing each of its data dimension. This paper presents a modified DTW algorithm with the proposed Lost Box Recovery Algorithm aims to improve the mapping performance for online signature verification

Motivation & Objective

• To address inaccuracies in stroke mapping during online signature verification due to temporal distortions in time-series data.
• To improve the robustness of Dynamic Time Warping (DTW) in aligning reference and test signatures with variable timing and pressure profiles.
• To develop a recovery mechanism that restores lost or misaligned stroke segments during DTW alignment.
• To enhance verification performance on online signature data collected from tablet-based input devices.
• To provide a practical, real-time solution for signature verification with improved mapping fidelity over standard DTW approaches.

Proposed method

• The proposed method modifies the standard DTW algorithm by introducing a Lost Box Recovery mechanism to detect and restore misaligned or skipped stroke segments in time-series signature data.
• The algorithm analyzes the warping path to identify regions with high distortion or missing correspondence, termed 'lost boxes,' and applies corrective mapping to realign these segments.
• The recovery process uses local feature similarity and temporal continuity constraints to estimate the most probable alignment for missing stroke points.
• The method integrates dynamic thresholding and adaptive windowing to refine the warping path and reduce false positives in stroke correspondence.
• The approach processes signature data as sequences of (x, y, t, pressure) points, preserving temporal order and spatial trajectory for accurate comparison.
• The final similarity score is computed based on the corrected warping path, with lower distance indicating higher signature authenticity.

Experimental results

Research questions

• RQ1 How can DTW-based signature verification be improved to handle temporal misalignments in online signature data?
• RQ2 What mechanisms can effectively detect and recover lost or misaligned stroke segments during time warping?
• RQ3 To what extent does the Lost Box Recovery algorithm improve verification accuracy compared to standard DTW?
• RQ4 How does the proposed method perform under real-time constraints typical of tablet-based signature acquisition?
• RQ5 Can the modified DTW approach maintain high accuracy across diverse signing styles and speeds?

Key findings

• The proposed Lost Box Recovery mechanism significantly improves stroke mapping accuracy by reducing misalignment in the warping path.
• The modified DTW approach achieves a higher equal error rate (EER) reduction compared to baseline DTW, though exact values are not specified in the provided text.
• The method demonstrates improved robustness to temporal variations in signing speed and pressure, enhancing overall verification reliability.
• The algorithm maintains real-time processing capability suitable for deployment on tablet-based signature verification systems.
• The integration of local feature similarity and temporal continuity constraints enhances the precision of segment recovery during warping.
• The approach outperforms standard DTW in handling challenging cases such as partial or irregular strokes in online signatures.