Abstract:
Signature veri cation is used in verifying the claimed identity of a person through
his/her chosen and previously registered signature. The signature's widespread acceptance
by the public and niche applications (validating paper documents and use
in banking applications) makes it a desirable biometric.
Signature is considered to be a behavioral biometric that encodes the ballistic movements
of the signer and as such is di cult to imitate. On the other hand, compared
to physical traits such as ngerprint, iris or face, a signature typically shows higher
intra-class and time variability. Furthermore, as with passwords, a user may choose
a simple signature that is easy to forge.
In this work, we present a state-of-the-art o ine signature veri cation system that
uses a fusion of complementary features, classi ers and preprocessing techniques, with
the aim to explore the limits in signature veri cation accuracy.
Here we propose an e cient way to verify the identity of a claimed person. This proposed
method uses graph matching for measuring the similarity between two sample
signatures. For making it eligible to apply graph matching, features are computed by
creating graph representation of the signature image. We propose to use direction information
for edge feature to make the method more robust to rotation of the image.
This direction is normalized to cope with the intra-class variability. For matching
Factorized graph matching is used as it provides faster computation of point-wise
correspondence. Based on the a nity score, a statistical approach is made for decision
making. For validation Writer-Dependent approach is applied.
Extensive experiments have been conducted on a benchmark and publicly available
O ine Signature Dataset (ICDAR 2011), where the proposed algorithm achieves approximately
12% average False Acceptance Rate (FAR) for signature veri cation. Furthermore,
comparison with other promising works on the aforementioned databases
demonstrates the enhanced performance and e ciency of the proposed method
