Online handwritten signature verification using feature weighting algorithm relief

Online handwritten signatures are widely used as a reliable identity authentication technology in industries such as banking, insurance and hospitals. Most existing online handwritten signature verification schemes usually choose the same feature sets for different users while comparing different signatures, which may reduce the stability of the dynamic characteristics. To avoid this problem, we propose a novel writer-dependent online signature verification technique based on Relief. For each user, our verification scheme falls into two phases: the training phase and the test phase. In the training phase, we select a signature as the base signature from real signatures and construct two different kinds of signature pairs: the real signature pairs and the forged signature pairs. Each pair of real signatures consists of two real signatures, and each pair of forgeries is made up of a real signature and a forgery. A group of combined features are defined for each pair of signatures. Then, we select more stable combined features by means of the Relief algorithm and compute a matrix used as a matching template, which contains the value of each combined feature for each pair of signatures. In the test phase, the classification of the pair of signatures composed of the test signature and the base reference signature is determined by the K-nearest neighbor. Experiments are conducted on our own database and SVC2004, and the results indicate that the method we propose here shows lower FAR and FRR than do traditional methods on the same databases.