Abstract
Abstract
This study investigated the discrimination between genuine and simulated digitally captured signatures (DCSs) to support their forensic examination. The dataset included genuine questioned signatures from 65 writers, collected across multiple sessions and in three writing positions, together with reference signatures collected in the standard position. In addition, eight simulations were produced by different individuals for each writer and treated as simulated questioned signatures, yielding a total of 520 simulations. Features of genuine and simulated questioned signatures were compared relative to writer‐specific reference material. All examined features except the size‐related characteristics and the number of pen lifts differed significantly between genuine and simulated signatures. Dynamic features, particularly total duration, achieved strong discrimination with an overall error rate as low as 7%. Combining features reduced the error rate marginally and not significantly (to 6%). Simulations were less often misclassified when reference signatures exhibited lower standardized within‐writer variability in duration. This suggests that correspondence between questioned and reference signatures may carry greater evidential value under such conditions. A similar, but non‐significant, trend was observed with increasing signature complexity. Exploratory analyses suggested benefits of writer‐specific interpretation of signature features. While no clear temporal effect on classification error rates was observed across the collection period, writing position influenced performance, highlighting the importance of considering non‐standard writing conditions in forensic practice. Overall, despite substantial variability in genuine signatures, DCS features enabled reliable discrimination between genuine and simulated signatures, with continuous dynamic features providing key discriminatory information and extending the possibilities of forensic examination beyond pen‐and‐paper signatures.