A novel adaptive two-phase multimodal biometric recognition system

Multimodal biometric recognition systems are intended to offer authentication without compromising on security, accuracy and these systems also used to address the limitations of unimodal systems like spoofing, intra class variations, noise and non-universality. In this paper, a novel adaptive two-phase multimodal framework is proposed with face, finger and speech traits. In this work, face trait reduces the search space by retrieving few possible nearest enrolled candidates to the probe using Gabor wavelets, semi-supervised kernel discriminant analysis and two dimensional- dynamic time warping. This nonlinear face classification serves as a search space reducer and affects the True Acceptance Rate (TAR). Later, level-1 and level-2 features of fingerprint trait are fused with Dempster Shafer theory and achieved high TAR. In the second phase, to reduce FAR and to validate the user identity, a text dependent speaker verification with RBFNN classifier is proposed. Classification accuracy of the proposed method is evaluated on own and standard datasets and experimental results clearly evident that proposed technique outperforms existing techniques in terms of search time, space and accuracy.