The importance of feature selection in online signature verification

Handwritten signatures are one of the most commonly used biometrics. Because signatures are widely accepted, their verification is a fundamental problem. Online signature verification applies different electronic devices to capture the signatures, which allow the use of dynamic information, such as the pressure or inclination angle of the pen. Therefore, it is much more challenging to forge online signatures. Further derived features can be defined and calculated based on the captured features. The selection of applied features is a crucial step of signature verification. This work aims to analyse the importance of feature selection by comparing the usability of common derived function features using dynamic time warping (DTW) based and Siamese bidirectional long short-term memory (BiLSTM)-based solutions. This paper comprehensively studies the most common function features and their literature background, presents several feature sets including pairs and triplets, compares their effectiveness using different classification approaches, and presents a DTW-based classifier and four Siamese BiLSTM models using two different loss functions.