Online Signature Verification Using Chi-Square (χ2) Feature Selection and SVM Classifier

Signature verification is a potential research area because of its social, legal and cultural acceptance since time immemorial. Hence unlike other biometrics it is more prone to forgery. So we have proposed an online signature verification system with user specific Chi-Square (χ2) feature selection to enhance the accuracy of our system. The aim of our work is to verify the signature of a person by comparing with the reference template for the claimed signature stored in the system during enrolment. Features extraction and feature selection are critical area because many number of features may be derived for verification purpose. So, to choose the most appropriate features describing a particular class is critical because it has a high effect on accuracy of the system. Therefore, we used the χ² feature selection algorithms to minimize verification error, decrease computational time, decrease model size and achieve higher performance. Signature verification is a very difficult work because of large intra-class variations and small interclass variations while taking into account skilled forgeries. Here the verification is done using skilled forged signatures of each users to enhance the verification accuracy. Efficiency of the model on unseen data is enhanced using k fold cross validation. In the proposed system we have used Multiple SVM with all the kernel functions to find the best performing SVM with the different numbers of Chi-Square (χ2) selected features to find out the model with most promising verification results. The Model is used on the SVC2004 Database (Task 1) yielding very good result. Accuracy achieved in our case is 92.75 %.