Online Signature Verification through Scaled Histogram of Oriented Gradients
Mpilo Mshengu, Mandlenkosi Victor Gwetu · 2019
Since signatures are an integral part of our everyday lives, that represent a legally binding seal of approval; it is essential for them to be verified accurately. Online signature verification is used to confirm an individual's identity based on realtime characteristics that are compiled during the act of signing. It is often used to reduce fraud in institutions such as banks and other credit service providers, since it offers a robust alternative to the traditional offline signature verification. This study aims to complement existing efforts in this field by employing Histogram of Oriented Gradients (HOGs) as a feature extraction mechanism that feeds into a template matching exercise based on Normalized Cross Correlation (NCC). The scale at which HOGs are applied is controlled in order to investigate the trade-off between granularity and verification effectiveness. Accuracies of 78% and 70% are obtained on the SVC2004 dataset for complete and half signature coverage respectively. This highlights the importance of global context when using realtime features, since it appears that useful information is lost when using separate HOGs for the first and second half of an online signature.