A unified authentication framework using finger nail plate biometric

This paper proposes a novel approach to segment the nail-plate as the region of interest (ROI) from the acquired hand samples. The proposed approach works at the pixel-level and classifies each pixel into either nail-plate or non nail-plate regions. The Gabor-mask is employed on the extracted nail-plate to obtain an accurate ROI irrespective of grown nail-plate (as in female samples) by minimizing the segmentation error. Both Wavelet and Independent component analysis (ICA) feature-descriptors that we have used here exploit the local texture and shape of the segmented nail-plate regions. Various score-level-fusion schemes are implemented to assert the veracity of the claimed identity. Receiver operating characteristics (ROC) shows that ring nail-plate performs the best among the three nail-plates with ICA yielding better results than Wavelet. The fusion of ICA and Wavelet features shows that the performance of product-rule is better than sum-rule at FAR=0.1%. To consolidate the ranked-identities from three nail-plates of each sample, the rank level fusion of Wavelet and ICA features shows that the rank-1 recognition rate of Borda-count and Logistic-regression method is better than Highest-rank. Thus, the rigorous experimental analysis on 180 users with 5 samples per user and three nail-plates (mainly of index, middle, and ring finger) of each sample demonstrates the utility of this new biometric identifier for several medical, surveillance, and forensic applications.