Real-time verification of handwritten signatures using a hybrid opto-electronical method

We propose a hybrid optoelectronic method for the fast elimination of forgeries in a preclassification stage of a handwritten signature verification system. Our method consists of the combination of several statistical classifiers. The first step is to transform the original signatures using the identity and the Gabor transforms. Then, the intercorrelations of the transformed signatures with a reference database leads to a characterization in a n-dimensional feature space. We reduce the number of features to the most relevant ones by principal component analysis (PCA). The discriminant function of the classifier is defined as being the dilated convex hull of the reference database projected in the reduced feature space. The Gabor transforms and the intercorrelations are optically implementable. With the current high-speed optical correlators, we can expect to preclassify an unknown signature in about 125 ms. The application and the data are briefly summarized, our method is detailed, the related hybrid optoelectronic system is presented and a simulation is realized in the case of one writer. The preliminary results show that the fusion of the decisions taken for each transformation enables us to increase the rejection rate of the forgeries, while accepting all the genuine signatures, from 57% up to 69%