Analysis of 'goat' within user population of an offline signature biometrics

Intra - user variability inherent in human handwritten signatures remains one of the main challenges for a robust biometrics signature based authentication system. The existence of a subset of users classified as `goats' in the Doddington's menagerie whose signature samples are highly inconsistent and often rejected by the biometrics system may degrade the system accuracy by contributing a large portion to the False Rejection Rate (FRR). However, little is known on the level of the intra user variability and percentage of the `goats' in the overall user population, which in turns remains the prime focus of this paper. An HMM-based computational approach is used to build the reference model and verify the authenticity of an input sample based on a series of a local feature extracted from signature images. Here, four different goat populations are identified for offline signature biometric system which is based on four different local features ( namely pixel density, centre of gravity, angle, and distance) and are analysed for their co-relationship. The overall analysis is carried out on Sigma database which is compiled to reflect the signatures of a target user population.