Online Handwritten Signature Verification based on FIR Filters using Discrete Fractional Cosine Transform

Abstract

Biometrics is being widely implemented in today’s society to deal with the security requirement issues. A biometric system can either do identification or verification task. In identification the biometric system can accept or reject identity of a person whereas verification authenticates the person’s claimed identity from database samples. Biometric verification technology can be divided into two branches: physiological verification and behavioural verification. Physiological recognition includes speech, fingerprint, iris and retinal recognition etc. Physiological recognition is used to authenticate or identify an individual whereas Behavioural recognition examines the mannerisms of an individual for example keyboard typing recognition, signature verification etc., these verification techniques analyse and recognize how an individual signs his signature or uses a keyboard etc. Signature verification is an intriguing intellectual challenge with many practical applications. It is one of the customary ways to verify person in many countries. However, it is known that signatures signed by same person are never precisely the same. It is widely and commonly accepted practice for authentication of an individual. Whereas off-line signature verification contributes very less to accurate identification, on-line signature verification has been successfully implemented in recent researches to achieve 80%-98% of accuracy. Various approaches have been used to implement biometric signature recognition some of which are dynamic time warping (DTW), Bayesian Learning, Hidden Markov model (HMM), Neural Networks, Support Vector machine (SVM) etc. In literature, many techniques have developed given to extract online signature features. The best field of signature verification is still under developing phase, many methodologies are yet to be explored. One such less explored methodology is based on Fractional Transform. Fractional Transform is generalization of classical transforms. This transform has an additional parameter which gives an extra degree of freedom. This field is still in development. An algorithm for online handwritten signature verification which is based on discrete fractional Cosine Transform (DFrCT) is used. The Experiments were performed on SVC iv 2004 and SUSIG databases. In this system ten features of tested signature are extracted. Then system is realized with the help of five FIR filters. After this feature vectors are calculated. Euclidean norm of difference between the feature vectors of test signatures and reference signatures is calculated. This Euclidean norm is compared with threshold to classify the signatures as genuine or forgery. The equal error rate (EER) is calculated to compare the efficiency of this method with the existing methods. It has been observed that the results of proposed DFrCT with ten features are better than discrete cosine transform (DCT) results.