Applying personalized weights to a feature set for on-line signature verification

Abstract

For signature verification, there can be a large number of useful features including pen-speed, pen-pressure, etc. However, all these features may not be useful or even be harmful for a given person when not enough number of training samples are allowed. Namely, for an optimal verification, each person requires a different subset of all the possible features. Finding an optimal subset for an individual needs too much combinatorial efforts. Instead, we propose an approach using all the same features for every person by assigning a different weight to each feature according to the nature of the person's signature.