Offline Handwritten Signature Verification using Siamese Network Model
Wanghui Xiao · 2024
Offline handwritten signatures constitute one of the most prevalent and universally recognized identifiers within the domains of biometrics and document forensics. They are routinely employed for daily attendance tracking, credit card authorization, and the authentication of business agreements to confirm individual identities. The task of offline signature verification remains formidable due to the intricacies involved in distinguishing subtle yet critical variations between authentic and expertly forged signatures. To tackle this issue, this paper introduces a two-stage Siamese network framework for offline handwritten signature verification. It incorporates an advanced spatial transformation module that autonomously emphasizes the features indicative of handwriting style for more accurately capturing the unique handwriting characteristics of target signature. Furthermore, the Focal loss function is adopted to address the significant imbalance between positive and negative signature samples prevalent in most contemporary datasets. Experimental results on four real-world handwritten signature datasets in various languages demonstrate that our proposed model surpasses state-of-the-art approaches in verification accuracy.