Signature Verification with Dynamic RBF Networks and Time Series Motifs
Abstract
This article presents a novel classification algorithm for (multivariate) time series. In a first step, so-called time series motifs, which represent characteristic subsequences of the time series, are extracted using extreme points. In a second step, the extracted motifs are used to train a dynamic radial basis function network (DRBF). Compared to a standard radial basis function network, this DRBF has the advantage, that not only similar motifs of the same class are detected but also sequences of these motifs. For performance evaluation, the proposed classification algorithm is applied to online signature verification. Our experiments show, that the presented DRBF based on time series motifs is capable of a very reliable authentication with an equal error rate of about 1.5%.
Loading...