An Active Search of Local Individualities for an Off-line Signature Verification

We propose a new verification method that extracts local individual writing habits from signatures. In registration process, correspondence estimation is done with the reference signatures and strong parts of individual writing habits are extracted in the plural. In verification process, a questioned signature is checked by using the local individualities that is extracted from reference signatures after correspondence estimation. As for the correspondence estimation multi-scale regularization is made use of, and genetic algorithm is used for the extraction of strong parts of individual writing habits. As indexes that show the strength of individual writing habit, we use magnitude of movement when it was written repeatedly and complexity of pen stroke. Result of average error rate was 2.63% when verification experiment was done toward Japanese signatures of 20 subjects (800 examples).