Statistical on-line signature verification using rotation-invariant dynamic descriptors

The main challenges of online signature verification framework can be summarized in 1) accuracy and speed of preprocessing step, 2) the number of extracted features and how informative they are, and 3) complexity of decision making step owing to the less number of input samples for training. In this paper, in order to simplify the preprocessing step, a set of rotation invariant descriptors is presented. The complexity of system because of high dimensionality of signature features is reduced by exploiting the statistical moments of the wavelet coefficients. To improve it further, suggested Fisher-Metric(FM) filtering develops a feature selection algorithm. In the end, taking advantage of artificial neural network on SVC2004 database blow up the performance of the proposed approach with False Acceptance Rate (FAR) of 3% and False Rejection Rate (FRR) of 2.5% with an equal error rate (ERR) of less than 3%. The validation of the proposed framework suggests that the proposed method outperforms the state-of-the-art techniques.