Performance Evaluation of the JK-nearest Neighbor Online Signature Verification Parameters
Mohammad Saleem, Bence Kővári · 2021
There are several classification algorithms used for signature verification purposes. The k-nearest neighbor (KNN) algorithm was previously used in online signature verification, but in this paper, we present an evaluation of the online signature verification using the JKNN classifier, which is a generalized case of the KNN classifier. An optimal classifier is used to evaluate the algorithm's main parameters that provide the most accurate verification results. The evaluation of each parameter of the algorithm is presented and tested using the SVC2004 database. Our results show that JKNN can provide accurate results by optimizing the main parameters of the algorithm. Both false acceptance rate and false rejection rate behave differently with different values of each parameter; the least average error rate can be achieved by choosing the parameters’ optimal values.