Abstract:
Among the various biometric modalities, signature verification remains one of the most
widely used methods to authenticate the identity of an individual. Signature verification,
the most important component of behavioral biometrics, has attracted significant research
attention over the last three decades. Despite extensive research, the problem still remains
open to research due to the variety of challenges it offers. The high intra-class variations
in signatures resulting from different physical or mental states of the signer, the
differences that appear with aging and the visual similarity in case of skilled forgeries etc.
are only a few to name. In our proposed work we have developed an efficient feature
analysis based system that distinguishes between genuine and forged signature. In this
research we have presented the evaluation of suitable features for offline signature
verification and a complete analysis is done on Local binary pattern descriptor,
Histogram of oriented gradient and Chain code feature. In our work in first step we have
extracted the LBP and HOG features and then in classification phase we apply Support
vector machine (SVM) as a classifier for decision whether the signature is genuine or
forged. Dutch offline signature verification data set is taken which is used in ICDAR
2013 competition. Total 648 signature samples are trained in our work. We have done
experiments on almost 20 authors in all. We have measured the performance evolution to
calculating FAR and FRR rate. In second step we have extracted the Differential chain
code feature and it combine with LBP and HOG feature and then SVM is trained
obtained the best accuracy rate. Each author has 24 genuine and 4 forged signatures in
case of combing these three features.
