Online signature verification using combination of two classifiers

The objective of signature verification is to distinguish forgery signature from genuine one. Online signature is the one which is registered through electronic devices such as digitizers and stored on computers in time sequence form. In this kind of signatures in addition to location information, time information such as speed and acceleration is stored. In this paper after accomplishment of some pre-processing procedures like normalization of signature size, smoothing and elimination of rotation on signatures using algorithms based on extremum matching of signals and ant colony, their time duration will be equalized. Afterwards, similarities between signatures will be determined using extended regression and finally will try to distinguish between forgery signatures from genuine one using support vector machine (SVM). The suggested online verification system is tested on SVC2004 signature set which is related to the first international signature verification competition and results are compared to respective results of participants. The results state that suggested method exhibits equal error rate (EER) of %4.3 in skilled forger group.