Improved Online Signature Verification System Using Discrete Fractional Cosine Transform

Abstract

The rapid development and increasing use of computer terminals, automatic tellers, and data banks has engendered the need to protect sensitive information. The Biometric technology uses behavioral characteristics of people such as fingerprints, face, iris, hand geometry, retina, voice etc. in an automated way to provide security while accessing computers and automated banking and personal identification. But the most popular modality is handwritten signature used in banking and commerce, transaction, credit card payments, cheque authentication for longer time. There are two ways for handwritten signature verification i.e. offline and online signature verification. In offline signature verification, signatures are scanned after it has been written. A more useful and attractive approach is to generate electric signals representative of the signature during the signing process known as online signature verification. The advantage of this approach is that signature verification is based on the dynamics of the signature, which are not visible and, therefore, are very difficult to forge or copy. In the present work, an online signature verification system based on FIR system is designed using discrete fractional cosine transform (DFrCT). The DFrCT has extra degree of freedom to extract the feature of test signatures. The test signatures are preprocessed and then eight features horizontal movement, vertical movement, areal velocity, displacement, velocity, direction change, area pressure, motion pressure are extracted using DFrCT. The impulse responses of four FIR systems are used to calculate Euclidean norm. The signature can be verified by evaluating the difference between the average of Euclidean norms of genuine and forgery signatures. The equal error rate (EER), false acceptance rate (FAR), false rejection ratio (FRR) is calculated to compare the efficiency of the present work with the previous work. In the present work, the minimum EER of 3.5%, minimum FAR of 1.5% and minimum FRR of 2% is achieved to verify above statement. However, the use of four FIR systems and extraction of eight features has increased the complexity. The accuracy and security can be increased by extracting more than eight features in future.