Automatic Handwriting Verification (AHV)

摘要

Over a four-month period, 5,220 signatures and 1,740 numeric 'Signatures' were collected from 59 subjects writing with the SRI pen. Twelve trained 'forgers' attempted 648 forgeries; they were given copies of the true signer signatures as well as information about how the SRI system works and what measures would be used to perform the signature verification. The forgers were also allowed to practice their forgeries and to view video tapes of the true signers writing their signatures. Data were analyzed to determine Type I/Type II error curves and average access time. For typical conditions, the True vs. Forgeries Type I/Type II equal-error rate was on the order of one percent, and the average access time, including the time to write the signature, was 8.5 seconds. These results are based on a features algorithm for signature verification using the same set of features for all subjects. It is shown that significant improvement may be achieved with the features technique if the feature sets are individualized for each user. Even greater improvement can be achieved with a correlation method of analysis, although this requires an increase in memory storage costs and computer processing time. (Author)