Handwritten Signature Verification System Using Sound as a Feature

Abstract

In this paper, the sound arising from the friction of paper and pen is considered as a biometric data for handwritten signature verification. Sounds were recorded with two different mobile phone models with different operating systems. To expand the scale of the study, the data collection process was carried out by taking samples from participants of different age ranges using different types of pen, paper, and mobile phones. Time sequential peak values were obtained from the onset strength envelopes of sound signals in the feature extraction phase. Similarity distances of the signatures were calculated by Dynamic Time Warping (DTW) algorithm. Equal Error Rate (EER) was used to measure the classification success. When calculated with signer-specific thresholds, EER values vary between %8.14 and %16.61. When calculated using a single threshold for all participant signers, EER values vary between %15.29 and %28.45 according to different pen, paper, and mobile phone combinations.