Combining Distances through an Auto-Encoder Network to Verify Signatures

In this paper we present a system for offline signature verification. The paperpsilas contributions are: i) Five distances were calculated and evaluated over the signature database, they are: furthest, nearest, template, central and n central. Also, a normalization procedure is established to turn each distance scale invariant; ii) These distances are combined using the following rules: product, mean, maximum and minimum; iii) The calculated distances can be used as a feature vector to represent a given signature. So,the feature vectors found and their combination were finally used as input vector for an auto-encoder neural network. All the experimental study is done using one-class classification, which demands only the genuine signature to generalize. The proposed approaches achieved very good rates for the signature verification task.