Learning of prototypes and decision boundaries for a verification problem having only positive samples

Abstract

This paper deals with learning of prototypes using only positive samples for on-line signature verification. In our learning method, the required number of prototypical signatures is determined by grouping similar individual signatures, and appropriate thresholds are adjusted to the sample distribution. This method is experimentally compared with a case that only one prototype is assigned to each signer with uniform threshold, and another case that only one prototype is assigned to each signer with adaptive threshold.