Optimized jk-nearest neighbor based online signature verification and evaluation of main parameters

Abstract

In this paper, we propose an enhanced $jk$-nearest neighbor ($jk$-NN) algorithm for online signature verification. The effect of its main parameters is evaluated and used to build an optimized system. The results show that the $jk$-NN classifier improves the verification accuracy by 0.73–10% as compared to a traditional one-class $k$-NN classifier. The algorithm achieved reasonable accuracy for different databases: a 3.93% average error rate when using the SVC2004, 2.6% for the MCYT-100, 1.75% for the SigComp'11, and 6% for the SigComp'15 databases. These results followed a state-of-the-art accuracy evaluation where both forged and genuine signatures were used in the training phase. Another scenario is also presented in this paper by using an optimized $jk$-NN algorithm that uses specifically chosen parameters and a procedure to pick the optimal value for $k$ using only the signer’s reference signatures to build a practical verification system for real-life scenarios where only these signatures are available. By applying the proposed algorithm, the average error rates that were achieved were 8% for SVC2004, 3.26% for MCYT-100, 13% for SigComp'15, and 2.22% for SigComp'11.