Biometric studies with hidden Markov model and its extension on short fixed-text input
Md Liakat Ali, John Vincent Monaco, Charles C. Tappert · 2017
The hidden Markov model (HMM) and its extensions have been applied in numerous scientific and engineering areas. In speech recognition, HMMs still outperform many other models. HMMs have also demonstrated significant performance in signature and gesture recognition. Nonetheless, the performance of HMM in Keystroke Biometric (KB) systems is typically lower when compared to other biometric systems. Moreover, there seems to be limited research conducted in Keystroke Biometrics using HMMs. This study discusses the hidden Markov model and its various extensions used in biometric areas. It also evaluates the hidden Markov model and the recently proposed partially observable hidden Markov model (POHMM) on a benchmark KB dataset. The POHMM, an extension of HMM that conditions the hidden state on an independent Markov chain, achieves a 0.045 equal-error rate (ERR), a significant performance improvement over the HMM and other leading methods in user verification.