Exploring Vision-Language Models for Online Signature Verification: A Zero-Shot Capability Study

2026-05-14

AI Summary

This work presents the first exploration of zero-shot applicability of vision-language models (VLMs) to online signature verification. By transforming temporal signature data into static images enriched with pressure information, the study leverages state-of-the-art VLMs such as GPT-5.2 and Gemini 2.5 Pro for zero-shot verification and introduces a biometric scoring mechanism based on token probabilities. Experimental results demonstrate that the proposed approach achieves an equal error rate as low as 0.32% under random forgery scenarios on mobile devices, outperforming current supervised state-of-the-art methods. However, performance degrades significantly in skilled forgery tasks, and chain-of-thought (CoT) reasoning induces a “rationalization trap,” adversely affecting verification accuracy. This study thus reveals both the promising potential and inherent limitations of VLMs in high-precision biometric recognition.

Abstract

Recent advancements in Vision-Language Models (VLMs) have demonstrated strong capabilities in general visual reasoning, yet their applicability to rigorous biometric tasks remains unexplored. This work presents an exploratory study evaluating the zero-shot performance of state-of-the-art VLMs (GPT-5.2 and Gemini 2.5 Pro) on the Signature Verification Challenge (SVC) benchmark. To enable visual processing, raw kinematic time-series are converted into static images, encoding pressure information into stroke opacity whenever available in the source data. Furthermore, we introduce a scoring protocol that extracts latent token probabilities to compute robust biometric scores. Experimental results reveal a significant performance dichotomy dependent on signal quality and forgery type. In random forgery scenarios, the zero-shot VLM achieves exceptional discrimination, with GPT-5.2 reaching an Equal Error Rate of 0.32% in mobile tasks, outperforming supervised state-of-the-art systems. Conversely, in skilled forgery scenarios, where the task is more challenging because both signatures are almost identical, the results are significantly worse, and a critical "Rationalization Trap" emerges: chain-of-thought (CoT) reasoning degrades performance as the model produces kinematic hallucinations to justify forgery artifacts as natural variability.

Problem

• Vision-Language Models
• Online Signature Verification
• Zero-Shot Learning
• Biometric Authentication
• Forgery Detection

Innovation

• Vision-Language Models
• Zero-Shot Learning
• Online Signature Verification
• Kinematic-to-Image Encoding
• Rationalization Trap