Automatic Signature Verification Based on Accelerometry

Abstract

The fine structure of the muscle forces that are exerted during the writing of a signature is constant and well defined for most people. In general, the fine structure is not subject to conscious control. Based on these observations, an experimental system has been designed that utilizes a person's signature dynamics to verify identities. The design and operational features of this system are described. Experiments on 70 subjects during a four-week period show a 2.9 percent rejection of valid signatures and a 2.1 percent acceptance of forgeries. An average of 1.2 trials was necessary for verification. The forgers were knowledgeable about the verification technique and did their best to deceive the system. The acceptance rate of random forgeries, i.e., accidental matching of two separate signatures, was 0.16 percent.