An HMM for online signature verification based on velocity and hand movement directions
Saeede Anbaee Farimani, Majid Vafaei Jahan · 2018
The online signature is a behavioural biometrie trait and forms time series for verifying a persons identity. In this paper, a new online signature verification system using Hidden Markov Model (HMM) is presented. Prior model-based approachs use the combination of HMM and Gaussian Mixture Model, which causes the problem of proper selection of HMM states numbers. The advantage of the proposed HMM is the small number of observations and proper definition of states for time series modelling. The proposed system segments each signature curve based on pen's velocity value. The signature curve, would be decomposed in low or high partition according to velocity's value. For each partition, hand movement direction between two consequent point Extracted. The resulted time sequence of directions is then used for training an HMM. Development and evaluation experiments are reported on SVC2004 dataset. Two partitioning methods based on velocity or pressure are examined. Experiment result suggest that our system, with velocity based partitioning perform better results. FAR = 4.8% and FRR = 5%.