Evaluating the Performance of Machine Learning Models in Handwritten Signature Verification

Abstract

Handwritten signatures remain a widely used method for personal authentication in various official documents, including bank checks and legal papers. The verification process is often labor-intensive and time-consuming, necessitating the development of efficient methods. This study evaluates the performance of machine learning models in handwritten signature verification using the ICDAR 2011 Signature and CEDAR datasets. The investigation involves preprocessing, feature extraction using CNN architectures, and optimization techniques. The most effective models undergo a rigorous evaluation process, followed by classification using supervised ML algorithms, such as linear SVM, random forest, logistic regression, and polynomial SVM. The results indicate that the VGG16 architecture, optimized with the Adam optimizer, achieves the satisfactory performance metrics. This study demonstrates the potential of ML methodologies to enhance the efficiency and accuracy of signature verification, offering a robust solution for document authentication.