An Efficient Approach for Recognition and Verification of On-Line Signatures Using PSO

Abstract

Signature recognition and verification have been widely used for user authentication. A person is allowed to proceed further only when his/her signature matches with his/her model or template(s) stored in the database. In this paper, a robust approach for online signature recognition and verification has been proposed. Signatures have been segmented into uniform segments and features are collected from each segment to capture local dynamic properties. For signature recognition task, Random Forest is used as the classifier and particle swarm optimization(PSO) has been used to select the best feature-set and model parameter. The feature set and parameters selected from recognition task have been used in training binary random forest classifiers for user verification. Signature verification has been performed in two modes i.e. using global threshold and local threshold and corresponding results have been reported. In our experiment, we have used two public datasets (MCYT-100 and SVC-2004) and have achieved over 99% recognition rate and encouraging Equal Error Rate (EER) for verification on both the datasets.