Several papers have addressed the problem of providing automatic verification of handwritten signatures, by a template-matching approach using signals generated by the motions of the pen. The use of dynamic information (rather than, say, an analysis of the static image) lends itself more readily to simple pattern recognition approaches and in addition provides a greater measure of protection against forgery.

A major problem however results from inconsistencies in the rate of signing, analogous to the nonlinear time-warping which occurs in vocal utterances. Attempts have been made to deal with this problem by means of a rather ad-hoc segmented-correlation approach.

Nonlinear time-warping by dynamic programming has proved highly successful in addressing this problem in the field of speech recognition, and we have tried applying it to signature verification. Due to the limited information content of a signature compared with a vocal utterance the computational burden is modest.

In addition, we have investigated the feasibility of performing the verification solely on the basis of the pen pressure, inferred by measuring the reaction force of the writing surface. This would allow signature verification to be carried out without the need for a special instrumented pen.

Preliminary results based on limited testing indicate that a success rate of 97% can be achieved, when the decision threshold is set to equalise false-acceptance/false rejection rates.