Countermatch: A Neural Network Approach to Automatic Signature Verification

Abstract

The COUNTERMATCH algorithm is an elastic matching neural network model which attempts to link all the corresponding points between two dynamic signatures; a signature is simply an ordered set of strokes, each stroke being an ordered list of inking pen positions derived from fixed period sampling on a graphics tablet. Once a point-for-point correspondence has been defined, the algorithm can assign a numerical dissimilarity between the two signatures on the basis of the amount of elastic deformation required to change one into the other. A simple (or complex) discriminant can then be constructed from a training set of dissimilarities derived from known authentic and forgery matches.