Writer specific parameters for online signature verification

Abstract

In this work, we present a new model which is capable of handling variations in signatures for better representation by employing structure preserving feature selection method and representing the selected data in the form of interval representation scheme. The proposed model represents each writer with writer dependent dimension and authentication threshold. Decisions on the number of features to be used for each writer and the similarity threshold for deciding the authenticity of a given signature are arrived based on minimum equal error rate (EER) criteria. Based on the symbolic representation, a method of verification is proposed. The proposed model is tested for its effectiveness on benchmarking MCYT (DB1) and MCYT (DB2) datasets consisting of signatures of 100 and 330 writers respectively. The obtained results indicate the effectiveness of the proposed model.