An improved decision criterion for genuine/forgery classification in on-line signature verification
Alisher Kholmatov and Berrin Yanikoglu
We present a system for on-line handwritten signature verification. During enrollment, eight reference signatures are taken from each subject and statistics describing the variation in the user's signatures are extracted from the cross alignment of these. A test signature's authenticity is established by first aligning it with each reference signature for the claimed user. Then, using the alignment scores, normalized by profile statistics as features, the test signature is classified into one of the two classes (genuine or forgery), using standard pattern classification techniques. We experimented with the Bayes classifier and a simple thresholding scheme after applying PCA to the alignment scores. Results were 1.4% equal error rate for a data set of 94 people and 495 test signatures (genuine and skilled forgeries).