Online Signature Verification Using Probablistic Modeling and Neural Network

Increasing needs for secure transaction processing using reliable methods makes the biometric overcome some of the limitations of the traditional personal identification technologies. An online signature is a behavioral biometric that still has some limitations to be applicable like other biometric identification because of its behavioral nature. So, new algorithms and solutions are still required. This paper presents a new technique by combining Back-propagation Neural Network (BPNN) technique and the probabilistic model to overcome some drawbacks of using a single model individually. The probabilistic model is used to classify the global features, while BPNN is used to classify the local features. "AND" fusion is used to combine the two mentioned techniques to obtain the final decision. The dataset used to test and evaluate the proposed method is the SVC2004 dataset which is a well known dataset. The proposed technique was evaluated in terms of False Rejection Rate (FRR) and False Acceptance Rate (FAR) that are 0.3% and 0.5% respectively. The results are very encouraging when compared with related existing studies.