Signature verification: Increasing performance by a multi-stage system

A serial three stage multi expert system for facing the problem of signatur e verification is proposed. The whole decision process is organised into su ccessive stages, each using a very reduced set of features for recognising, a forgeries and providing information about the reliability of the recognit ion process. The first expert, adopting only a single global feature, is de voted to the elimination of random and simple forgeries. The second stage r eceives only those signatures not classified as false by the first stage (i .e. those signatures really genuine or forgeries reproduced in a skilled wa y), and adopts a single specific feature suitable for isolating skilled for geries. Both of these two stages employ suitable criteria for estimating th e reliability of the performed classification, so that, in case of uncertai nty, the signature is forwarded to a final stage which takes the final deci sion, taking into account the decisions of the previous stages together wit h the corresponding reliability estimations. The proposed multi-stage autom atic signature verification system has been tested on a database of signatu res produced by 49 different writers. The experimental analysis highlights the effectiveness of the approach: the proposed system employing only two f eatures, used in distinct moments of the decision process, performs better than other systems, employing larger feature set (including the features us ed in the proposed system) and performing classification in a single stage.