Scholarlink

<p>Signature verification is a critical biometric authentication task, challenged by high intra-class variations and skilled forgeries. This paper presents M2SM, a multi-modal signature matching framework that leverages a novel metric learning curriculum to enhance online signature verification. Unlike conventional methods, M2SM represents signatures across three modalities: sequential frames (video), image, and points-to capture comprehensive spatio-temporal features. A curriculum-based learning strategy dynamically adjusts triplet selection to optimize contrastive loss, ensuring robust intra-class discrimination. We evaluate M2SM on publicly available datasets, achieving an Equal Error Rate (EER) of 1.59% for skilled forgeries and 0.76% for random forgeries, outperforming state-of-the-art approaches.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

M2SM: multi modal signature matching network utilizing spatio-temporal features extracted from online signature

  • 期刊论文

  • 发表日期：

    2025年4月4日

查看全文
  • Anurag Pandey,
  • Pushap Deep Singh,
  • Arnav Bhavsar Vinayak,
  • Aditya Nigam,
  • Divya Acharya,
  • Prabhishek Singh,
  • Arpit Bhardwaj,
  • Manoj Diwakar

摘要

Signature verification is a critical biometric authentication task, challenged by high intra-class variations and skilled forgeries. This paper presents M2SM, a multi-modal signature matching framework that leverages a novel metric learning curriculum to enhance online signature verification. Unlike conventional methods, M2SM represents signatures across three modalities: sequential frames (video), image, and points-to capture comprehensive spatio-temporal features. A curriculum-based learning strategy dynamically adjusts triplet selection to optimize contrastive loss, ensuring robust intra-class discrimination. We evaluate M2SM on publicly available datasets, achieving an Equal Error Rate (EER) of 1.59% for skilled forgeries and 0.76% for random forgeries, outperforming state-of-the-art approaches.