Model Performance Comparison and Evaluation of Handwritten Signature Recognition
Zhilu Yang · 2024
In the digital era, the importance of information security remains a crucial and constantly relevant topic. Despite the development of various biometric features over the years to ensure the security of online transactions and authentication, biometric verification based on handwritten signatures remains the most popular method in official security protocols. It plays an irreplaceable role in societal life and information security. Therefore, our experiment focused on a moderately sized dataset of handwritten signature images with a certain level of complexity and diversity. The study involved the training and prediction of signature recognition models using Capsule Network and Support Vector Machine (SVM), with a quantitative evaluation of their performance. The results demonstrated that, compared to the SVM model, Capsule Network exhibited superior performance in terms of accuracy, precision, recall, and F1 score, with improvements of 19.47%, 8.78%, 24.55%, and 16.67%, respectively. This research provides a comprehensive analysis of the performance comparison between Capsule Network and SVM in the field of handwritten signature fraud detection, offering valuable insights for enhancing identity verification security.