Optimum Feature Selection for Signature Verification Using Different Classifiers

Abstract

Signature verification is an important parameter in the biometrics area as it is one of the behavioral biometric traits, which are widely used as a means of personal verification. Therefore, they require efficient and meticulous methods of authenticating the users. Features used in this work are employed for performance analysis using different combinations among them. For optimum feature selection Fisher‘s Discrimination ratio (FDR) is calculated for each individual feature and according to the calculated FDR, best feature is identified. Three classifiers: k-Nearest Neighbor (k-NN) classifier, Random Forest classifier and Naïve bayes classifier are used to verify whether the signatures are real or forgery. The results of all the classifiers are compared.