Present-day off-line signature verification methods definitely could and should be
improved, considering that not even the best systems can achieve lower error rates
than 5 percent. In this paper we present an off-line comparison method for differentiating
between genuine and forged signatures based on feature matching, specifically baseline
matching. Since a highly modularized framework has already been created, we developed
different modules that suited that system, and were able to create a module chain
that extracted baseline information from the signatures, and using the knowledge gained
from a small learning set could decide whether the signature was forged or genuine.
Of course verification based on only one feature can not be perfect, but the results
imply that involving additional - more or less - independent features of the signature
can decrease the error rate of the system below the barrier of 5 percent.