Structural hidden Markov models for biometrics: Fusion of face and fingerprint

The goal of this paper is threefold: (i) propose a novel face and fingerprint feature modeling using the structural hidden Markov models (SHMMs) paradigm, (ii) explore the use of some feature extraction techniques such as ridgelet transform, discrete wavelet transform with various classifiers for biometric identification, and (iii) determine the best method for classifier combination. The experimental results reported in both fingerprint and face recognition reveal that the SHMMs concept is promising since it has outperformed several state-of-the-arts classifiers when combined with the discrete wavelet transform. Besides, this study has shown that the ridgelet transform without principal components analysis (PCA) dimension reduction fits better with the support vector machines (SVMs) classifier than it does with the SHMMs in the fingerprint recognition task. Finally, these results also reveal a small improvement of the bimodal biometric system over unimodal systems; which suggest that a most effective fusion scheme is necessary.